[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <15ec9cce-8cce-0bfb-44d2-1f8e5acaebc0@partner.samsung.com>
Date: Wed, 17 Apr 2019 13:00:05 +0200
From: Lukasz Luba <l.luba@...tner.samsung.com>
To: Chanwoo Choi <cw00.choi@...sung.com>, devicetree@...r.kernel.org,
linux-kernel@...r.kernel.org, linux-pm@...r.kernel.org,
linux-samsung-soc@...r.kernel.org,
linux-arm-kernel@...ts.infradead.org
Cc: b.zolnierkie@...sung.com, krzk@...nel.org, robh+dt@...nel.org,
mark.rutland@....com, kyungmin.park@...sung.com,
m.szyprowski@...sung.com, s.nawrocki@...sung.com,
myungjoo.ham@...sung.com, kgene@...nel.org
Subject: Re: [PATCH 2/3] drivers: devfreq: events: extend events by type of
counted data
Hi Chanwoo,
On 4/17/19 10:22 AM, Chanwoo Choi wrote:
> Hi Lukasz,
>
> On 19. 4. 17. 오후 4:48, Lukasz Luba wrote:
>> This patch adds posibility to choose what type of data should be counted
>> by the PPMU counter. Now the type comes from DT where the event has been
>> defined. When there is no 'event-data-type' the default value is used,
>> which is 'read data in bytes'.
>> It is needed when you want to know not only read+write data bytes but
>> i.e. only write data in byte, or number of read requests, etc.
>>
>> Signed-off-by: Lukasz Luba <l.luba@...tner.samsung.com>
>> ---
>> drivers/devfreq/event/exynos-ppmu.c | 38 +++++++++++++++++++++++++------------
>> include/linux/devfreq-event.h | 6 ++++++
>> 2 files changed, 32 insertions(+), 12 deletions(-)
>>
>> diff --git a/drivers/devfreq/event/exynos-ppmu.c b/drivers/devfreq/event/exynos-ppmu.c
>> index c61de0b..e5f81a5 100644
>> --- a/drivers/devfreq/event/exynos-ppmu.c
>> +++ b/drivers/devfreq/event/exynos-ppmu.c
>> @@ -140,6 +140,7 @@ static int exynos_ppmu_set_event(struct devfreq_event_dev *edev)
>> int id = exynos_ppmu_find_ppmu_id(edev);
>> int ret;
>> u32 pmnc, cntens;
>> + u32 evt_sel = edev->desc->data_type;
>>
>> if (id < 0)
>> return id;
>> @@ -154,9 +155,12 @@ static int exynos_ppmu_set_event(struct devfreq_event_dev *edev)
>> if (ret < 0)
>> return ret;
>>
>> - /* Set the event of Read/Write data count */
>> - ret = regmap_write(info->regmap, PPMU_BEVTxSEL(id),
>> - PPMU_RO_DATA_CNT | PPMU_WO_DATA_CNT);
>> + /* Check if the data type has been defined in DT, use default if not */
>> + if (evt_sel == UINT_MAX)
>> + evt_sel = PPMU_RO_DATA_CNT | PPMU_WO_DATA_CNT;
>> +
>> + /* Set the event of Read or Write or both (RDWR) data count */
>> + ret = regmap_write(info->regmap, PPMU_BEVTxSEL(id), evt_sel);
>> if (ret < 0)
>> return ret;
>>
>> @@ -357,6 +361,7 @@ static int exynos_ppmu_v2_set_event(struct devfreq_event_dev *edev)
>> unsigned int pmnc, cntens;
>> int id = exynos_ppmu_find_ppmu_id(edev);
>> int ret;
>> + u32 evt_sel = edev->desc->data_type;
>>
>> /* Enable all counters */
>> ret = regmap_read(info->regmap, PPMU_V2_CNTENS, &cntens);
>> @@ -368,24 +373,28 @@ static int exynos_ppmu_v2_set_event(struct devfreq_event_dev *edev)
>> if (ret < 0)
>> return ret;
>>
>> - /* Set the event of Read/Write data count */
>> +
>> + /* Set the event of proper data type counting.
>> + * Check if the data type has been defined in DT,
>> + * use default if not.
>> + */
>> switch (id) {
>> case PPMU_PMNCNT0:
>> case PPMU_PMNCNT1:
>> case PPMU_PMNCNT2:
>> - ret = regmap_write(info->regmap, PPMU_V2_CH_EVx_TYPE(id),
>> - PPMU_V2_RO_DATA_CNT | PPMU_V2_WO_DATA_CNT);
>> - if (ret < 0)
>> - return ret;
>> + if (evt_sel == UINT_MAX)
>> + evt_sel = PPMU_V2_RO_DATA_CNT | PPMU_V2_WO_DATA_CNT;
>> break;
>> case PPMU_PMNCNT3:
>> - ret = regmap_write(info->regmap, PPMU_V2_CH_EVx_TYPE(id),
>> - PPMU_V2_EVT3_RW_DATA_CNT);
>> - if (ret < 0)
>> - return ret;
>> + if (evt_sel == UINT_MAX)
>> + evt_sel = PPMU_V2_EVT3_RW_DATA_CNT;
>> break;
>> }
>>
>> + ret = regmap_write(info->regmap, PPMU_V2_CH_EVx_TYPE(id), evt_sel);
>> + if (ret < 0)
>> + return ret;
>> +
>> /* Reset cycle counter/performance counter and enable PPMU */
>> ret = regmap_read(info->regmap, PPMU_V2_PMNC, &pmnc);
>> if (ret < 0)
>> @@ -508,6 +517,7 @@ static int of_get_devfreq_events(struct device_node *np,
>> struct device *dev = info->dev;
>> struct device_node *events_np, *node;
>> int i, j, count;
>> + int ret;
>>
>> events_np = of_get_child_by_name(np, "events");
>> if (!events_np) {
>> @@ -544,6 +554,10 @@ static int of_get_devfreq_events(struct device_node *np,
>> desc[j].driver_data = info;
>>
>> of_property_read_string(node, "event-name", &desc[j].name);
>> + ret = of_property_read_u32(node, "event-data-type",
>> + &desc[j].data_type);
>> + if (!ret)
>> + desc[i].data_type = UINT_MAX;
>
> I think that it is not proper to use 'UINT_MAX'.
> If you want to use the default value, you can initialize
> the 'desc[i].data_type' according to the device type or compatible name.
>
> For all exynos SoC except for exynos5433,
> desc[i].data_type = PPMU_RO_DATA_CNT | PPMU_WO_DATA_CNT;
>
> For exynos5433,
> desc[i].data_type = PPMU_V2_EVT3_RW_DATA_CNT;
>
OK, I will set these default values here.
>>
>> j++;
>> }
>> diff --git a/include/linux/devfreq-event.h b/include/linux/devfreq-event.h
>> index 4db00b0..cc160b1 100644
>> --- a/include/linux/devfreq-event.h
>> +++ b/include/linux/devfreq-event.h
>> @@ -81,14 +81,20 @@ struct devfreq_event_ops {
>> * struct devfreq_event_desc - the descriptor of devfreq-event device
>> *
>> * @name : the name of devfreq-event device.
>> + * @data_type : the data type which is going to be counted in the register.
>> * @driver_data : the private data for devfreq-event driver.
>> * @ops : the operation to control devfreq-event device.
>> *
>> * Each devfreq-event device is described with a this structure.
>> * This structure contains the various data for devfreq-event device.
>> + * The data_type describes what is going to be counted in the register.
>> + * It might choose to count e.g. read requests, write data in bytes, etc.
>> + * The full supported list of types is present in specyfic header in:
>> + * include/dt-bindings/pmu/.
>
> Why did you make the path name as the 'pmu' instead of 'ppmu'?
The directory may contain in future other Performance Monitoring Unit
subsystems. In Exynos case we have: Platform Performance Monitoring Unit
but different SoCs might have other names, like: System PMU.
Since there is no 'pmu' in this directory, I am allocating it for
Performance Monitoring Unit and not Power Management Unit.
>
> It might have some confusion because 'pmu' has
> the two meaning in linux kernel.
> - PMU, Power Management Unit
> - PMU, Performance Monitoring Unit
Regards,
Lukasz
>
>> */
>> struct devfreq_event_desc {
>> const char *name;
>> + u32 data_type;
>> void *driver_data;
>>
>> const struct devfreq_event_ops *ops;
>>
>
>
Powered by blists - more mailing lists