Linux-Samsung-soc Archive on lore.kernel.org
 help / color / Atom feed
* [RFC PATCH 0/2] PM / devfreq: Add delayed timer for polling
       [not found] <CGME20200703061508epcas1p171aa3c0ab832b77e5837d8bd1e563742@epcas1p1.samsung.com>
@ 2020-07-03  6:26 ` Chanwoo Choi
       [not found]   ` <CGME20200703061509epcas1p10446471df6497b4b2d7066280daa7f87@epcas1p1.samsung.com>
                     ` (4 more replies)
  0 siblings, 5 replies; 11+ messages in thread
From: Chanwoo Choi @ 2020-07-03  6:26 UTC (permalink / raw)
  To: lukasz.luba, k.konieczny, krzk, kgene
  Cc: s.nawrocki, willy.mh.wolff.ml, b.zolnierkie, cw00.choi, chanwoo,
	myungjoo.ham, kyungmin.park, linux-pm, linux-kernel,
	linux-samsung-soc, linux-arm-kernel

Add the delayed timer to devfreq framework in order to support
the periodical polling mode without stop caused by CPU idle state.
Some Non-CPU device must need to monitor the device status like
utilization regardless of CPU state.

- patch1 explains the detailed reason why the delayed timer is required.
- patch2 initializes that exynos5422-dmc device use delayed timer as default
instead of deferrable timer.

Chanwoo Choi (2):
  PM / devfreq: Add support delayed timer for polling mode
  memory: samsung: exynos5422-dmc: Use delayed timer as default

 Documentation/ABI/testing/sysfs-class-devfreq | 12 +++
 drivers/devfreq/devfreq.c                     | 83 ++++++++++++++++++-
 drivers/memory/samsung/exynos5422-dmc.c       |  1 +
 include/linux/devfreq.h                       |  9 ++
 4 files changed, 104 insertions(+), 1 deletion(-)

-- 
2.17.1


^ permalink raw reply	[flat|nested] 11+ messages in thread

* [RFC PATCH 1/2] PM / devfreq: Add support delayed timer for polling mode
       [not found]   ` <CGME20200703061509epcas1p10446471df6497b4b2d7066280daa7f87@epcas1p1.samsung.com>
@ 2020-07-03  6:26     ` Chanwoo Choi
  0 siblings, 0 replies; 11+ messages in thread
From: Chanwoo Choi @ 2020-07-03  6:26 UTC (permalink / raw)
  To: lukasz.luba, k.konieczny, krzk, kgene
  Cc: s.nawrocki, willy.mh.wolff.ml, b.zolnierkie, cw00.choi, chanwoo,
	myungjoo.ham, kyungmin.park, linux-pm, linux-kernel,
	linux-samsung-soc, linux-arm-kernel

Until now, the devfreq driver using polling mode like simple_ondemand
governor have used only deferrable timer for reduing the redundant
power consumption. It reduces the CPU wake-up from idle due to polling mode
which check the status of Non-CPU device.

But, it has a problem for Non-CPU device like DMC device with DMA operation.
Some Non-CPU device need to do monitor continously regardless of CPU state
in order to decide the proper next status of Non-CPU device.

So, add support the dealyed timer for polling mode to support
the periodical monitoring. The devfreq driver and user can select
the kind of timer on either deferrable and delayed timer.

For example, change the timer type of DMC device
based on Exynos5422-based Odroid-XU3 as following:

- If want to use deferrable timer as following:
echo deferrable > /sys/class/devfreq/10c20000.memory-controller/timer

- If want to use delayed timer as following:
echo delayed > /sys/class/devfreq/10c20000.memory-controller/timer

Suggested-by: Lukasz Luba <lukasz.luba@arm.com>
Signed-off-by: Chanwoo Choi <cw00.choi@samsung.com>
---
 Documentation/ABI/testing/sysfs-class-devfreq | 12 +++
 drivers/devfreq/devfreq.c                     | 83 ++++++++++++++++++-
 include/linux/devfreq.h                       |  9 ++
 3 files changed, 103 insertions(+), 1 deletion(-)

diff --git a/Documentation/ABI/testing/sysfs-class-devfreq b/Documentation/ABI/testing/sysfs-class-devfreq
index 9758eb85ade3..b10aa3aa230d 100644
--- a/Documentation/ABI/testing/sysfs-class-devfreq
+++ b/Documentation/ABI/testing/sysfs-class-devfreq
@@ -65,6 +65,18 @@ Description:
 		as following:
 			echo 0 > /sys/class/devfreq/.../trans_stat
 
+What:		/sys/class/devfreq/.../timer
+Date:		July 2020
+Contact:	Chanwoo Choi <cw00.choi@samsung.com>
+Description:
+		This ABI shows and stores the kind of work timer by users.
+		This work timer is used by devfreq workqueue in order to
+		monitor the device status such as utilization. The user
+		can change the work timer on runtime according to their demand
+		as following:
+			echo deferrable > /sys/class/devfreq/.../timer
+			echo delayed > /sys/class/devfreq/.../timer
+
 What:		/sys/class/devfreq/.../userspace/set_freq
 Date:		September 2011
 Contact:	MyungJoo Ham <myungjoo.ham@samsung.com>
diff --git a/drivers/devfreq/devfreq.c b/drivers/devfreq/devfreq.c
index 52b9c3e141f3..2a4aa8742520 100644
--- a/drivers/devfreq/devfreq.c
+++ b/drivers/devfreq/devfreq.c
@@ -49,6 +49,11 @@ static LIST_HEAD(devfreq_governor_list);
 static LIST_HEAD(devfreq_list);
 static DEFINE_MUTEX(devfreq_list_lock);
 
+static const char timer_name[][DEVFREQ_NAME_LEN] = {
+	[DEVFREQ_TIMER_DEFERRABLE] = { "deferrable" },
+	[DEVFREQ_TIMER_DELAYED] = { "delayed" },
+};
+
 /**
  * find_device_devfreq() - find devfreq struct using device pointer
  * @dev:	device pointer used to lookup device devfreq.
@@ -454,7 +459,17 @@ void devfreq_monitor_start(struct devfreq *devfreq)
 	if (devfreq->governor->interrupt_driven)
 		return;
 
-	INIT_DEFERRABLE_WORK(&devfreq->work, devfreq_monitor);
+	switch (devfreq->profile->timer) {
+	case DEVFREQ_TIMER_DEFERRABLE:
+		INIT_DEFERRABLE_WORK(&devfreq->work, devfreq_monitor);
+		break;
+	case DEVFREQ_TIMER_DELAYED:
+		INIT_DELAYED_WORK(&devfreq->work, devfreq_monitor);
+		break;
+	default:
+		return;
+	}
+
 	if (devfreq->profile->polling_ms)
 		queue_delayed_work(devfreq_wq, &devfreq->work,
 			msecs_to_jiffies(devfreq->profile->polling_ms));
@@ -771,6 +786,11 @@ struct devfreq *devfreq_add_device(struct device *dev,
 	devfreq->data = data;
 	devfreq->nb.notifier_call = devfreq_notifier_call;
 
+	if (devfreq->profile->timer < 0
+		|| devfreq->profile->timer >= DEVFREQ_TIMER_NUM) {
+		goto err_out;
+	}
+
 	if (!devfreq->profile->max_state && !devfreq->profile->freq_table) {
 		mutex_unlock(&devfreq->lock);
 		err = set_freq_table(devfreq);
@@ -1625,6 +1645,66 @@ static ssize_t trans_stat_store(struct device *dev,
 }
 static DEVICE_ATTR_RW(trans_stat);
 
+static ssize_t timer_show(struct device *dev,
+			     struct device_attribute *attr, char *buf)
+{
+	struct devfreq *df = to_devfreq(dev);
+
+	if (!df->governor)
+		return -EINVAL;
+
+	return sprintf(buf, "%s\n", timer_name[df->profile->timer]);
+}
+
+static ssize_t timer_store(struct device *dev, struct device_attribute *attr,
+			      const char *buf, size_t count)
+{
+	struct devfreq *df = to_devfreq(dev);
+	char str_timer[DEVFREQ_NAME_LEN + 1];
+	int timer = -1;
+	int ret = 0, i;
+
+	ret = sscanf(buf, "%16s", str_timer);
+	if (ret != 1)
+		return -EINVAL;
+
+	for (i = 0; i < DEVFREQ_TIMER_NUM; i++) {
+		if (!strncmp(timer_name[i], str_timer, DEVFREQ_NAME_LEN)) {
+			timer = i;
+			break;
+		}
+	}
+
+	if (timer < 0) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	if (df->profile->timer == timer) {
+		ret = 0;
+		goto out;
+	}
+
+	mutex_lock(&df->lock);
+	df->profile->timer = timer;
+	mutex_unlock(&df->lock);
+
+	ret = df->governor->event_handler(df, DEVFREQ_GOV_STOP, NULL);
+	if (ret) {
+		dev_warn(dev, "%s: Governor %s not stopped(%d)\n",
+			 __func__, df->governor->name, ret);
+		goto out;
+	}
+
+	ret = df->governor->event_handler(df, DEVFREQ_GOV_START, NULL);
+	if (ret)
+		dev_warn(dev, "%s: Governor %s not started(%d)\n",
+			 __func__, df->governor->name, ret);
+out:
+	return ret ? ret : count;
+}
+static DEVICE_ATTR_RW(timer);
+
 static struct attribute *devfreq_attrs[] = {
 	&dev_attr_name.attr,
 	&dev_attr_governor.attr,
@@ -1636,6 +1716,7 @@ static struct attribute *devfreq_attrs[] = {
 	&dev_attr_min_freq.attr,
 	&dev_attr_max_freq.attr,
 	&dev_attr_trans_stat.attr,
+	&dev_attr_timer.attr,
 	NULL,
 };
 ATTRIBUTE_GROUPS(devfreq);
diff --git a/include/linux/devfreq.h b/include/linux/devfreq.h
index 57e871a559a9..12782fbb4c25 100644
--- a/include/linux/devfreq.h
+++ b/include/linux/devfreq.h
@@ -31,6 +31,13 @@
 #define	DEVFREQ_PRECHANGE		(0)
 #define DEVFREQ_POSTCHANGE		(1)
 
+/* DEVFREQ work timers */
+enum devfreq_timer {
+	DEVFREQ_TIMER_DEFERRABLE = 0,
+	DEVFREQ_TIMER_DELAYED,
+	DEVFREQ_TIMER_NUM,
+};
+
 struct devfreq;
 struct devfreq_governor;
 
@@ -70,6 +77,7 @@ struct devfreq_dev_status {
  * @initial_freq:	The operating frequency when devfreq_add_device() is
  *			called.
  * @polling_ms:		The polling interval in ms. 0 disables polling.
+ * @timer:		Timer type is either deferrable or delayed timer.
  * @target:		The device should set its operating frequency at
  *			freq or lowest-upper-than-freq value. If freq is
  *			higher than any operable frequency, set maximum.
@@ -96,6 +104,7 @@ struct devfreq_dev_status {
 struct devfreq_dev_profile {
 	unsigned long initial_freq;
 	unsigned int polling_ms;
+	enum devfreq_timer timer;
 
 	int (*target)(struct device *dev, unsigned long *freq, u32 flags);
 	int (*get_dev_status)(struct device *dev,
-- 
2.17.1


^ permalink raw reply	[flat|nested] 11+ messages in thread

* [RFC PATCH 2/2] memory: samsung: exynos5422-dmc: Use delayed timer as default
       [not found]   ` <CGME20200703061509epcas1p3e338b614c44f5d8b566dae92f11cc8cb@epcas1p3.samsung.com>
@ 2020-07-03  6:26     ` Chanwoo Choi
  2020-07-08 12:33       ` Krzysztof Kozlowski
  0 siblings, 1 reply; 11+ messages in thread
From: Chanwoo Choi @ 2020-07-03  6:26 UTC (permalink / raw)
  To: lukasz.luba, k.konieczny, krzk, kgene
  Cc: s.nawrocki, willy.mh.wolff.ml, b.zolnierkie, cw00.choi, chanwoo,
	myungjoo.ham, kyungmin.park, linux-pm, linux-kernel,
	linux-samsung-soc, linux-arm-kernel

Use delayed timer as default instead of deferrable timer
in order to monitor the DMC status regardless of CPU idle.

Signed-off-by: Chanwoo Choi <cw00.choi@samsung.com>
---
 drivers/memory/samsung/exynos5422-dmc.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/memory/samsung/exynos5422-dmc.c b/drivers/memory/samsung/exynos5422-dmc.c
index 25196d6268e2..93e9c2429c0d 100644
--- a/drivers/memory/samsung/exynos5422-dmc.c
+++ b/drivers/memory/samsung/exynos5422-dmc.c
@@ -945,6 +945,7 @@ static int exynos5_dmc_get_cur_freq(struct device *dev, unsigned long *freq)
  * It provides to the devfreq framework needed functions and polling period.
  */
 static struct devfreq_dev_profile exynos5_dmc_df_profile = {
+	.timer = DEVFREQ_TIMER_DELAYED,
 	.target = exynos5_dmc_target,
 	.get_dev_status = exynos5_dmc_get_status,
 	.get_cur_freq = exynos5_dmc_get_cur_freq,
-- 
2.17.1


^ permalink raw reply	[flat|nested] 11+ messages in thread

* Re: [RFC PATCH 0/2] PM / devfreq: Add delayed timer for polling
  2020-07-03  6:26 ` [RFC PATCH 0/2] PM / devfreq: Add delayed timer for polling Chanwoo Choi
       [not found]   ` <CGME20200703061509epcas1p10446471df6497b4b2d7066280daa7f87@epcas1p1.samsung.com>
       [not found]   ` <CGME20200703061509epcas1p3e338b614c44f5d8b566dae92f11cc8cb@epcas1p3.samsung.com>
@ 2020-07-03 12:33   ` Willy Wolff
  2020-07-08 14:25     ` Lukasz Luba
  2020-07-08 10:52   ` Bartlomiej Zolnierkiewicz
  2020-07-08 14:01   ` Lukasz Luba
  4 siblings, 1 reply; 11+ messages in thread
From: Willy Wolff @ 2020-07-03 12:33 UTC (permalink / raw)
  To: Chanwoo Choi
  Cc: lukasz.luba, k.konieczny, krzk, kgene, s.nawrocki, b.zolnierkie,
	chanwoo, myungjoo.ham, kyungmin.park, linux-pm, linux-kernel,
	linux-samsung-soc, linux-arm-kernel

Hi Chanwoo,

I think it doesn't help on the benchmark I suggested that is doing only memory
accesses. With both timer, I have the same timing.

To test the benchmark with these new patches about timer:

git clone https://github.com/wwilly/benchmark.git \
  && cd benchmark \
  && source env.sh \
  && ./bench_build.sh \
  && bash source/scripts/test_dvfs_mem_patched.sh

The benchmark is set by default to run for 1s, but you can increase this by
tweaking the script as:

taskset 8 ./bench_install/bin/microbe_cache 33554431 0 9722222 <TIME in sec> ${little_freq}


Also, as I reported the issue, would it be possible to add a
Reported-by: Willy Wolff <willy.mh.wolff.ml@gmail.com> ?
Many thanks in advance.


Best Regards,
Willy

^ permalink raw reply	[flat|nested] 11+ messages in thread

* Re: [RFC PATCH 0/2] PM / devfreq: Add delayed timer for polling
  2020-07-03  6:26 ` [RFC PATCH 0/2] PM / devfreq: Add delayed timer for polling Chanwoo Choi
                     ` (2 preceding siblings ...)
  2020-07-03 12:33   ` [RFC PATCH 0/2] PM / devfreq: Add delayed timer for polling Willy Wolff
@ 2020-07-08 10:52   ` Bartlomiej Zolnierkiewicz
  2020-07-08 14:01   ` Lukasz Luba
  4 siblings, 0 replies; 11+ messages in thread
From: Bartlomiej Zolnierkiewicz @ 2020-07-08 10:52 UTC (permalink / raw)
  To: Chanwoo Choi
  Cc: lukasz.luba, k.konieczny, krzk, kgene, s.nawrocki,
	willy.mh.wolff.ml, chanwoo, myungjoo.ham, kyungmin.park,
	linux-pm, linux-kernel, linux-samsung-soc, linux-arm-kernel


Hi Chanwoo,

On 7/3/20 8:26 AM, Chanwoo Choi wrote:
> Add the delayed timer to devfreq framework in order to support
> the periodical polling mode without stop caused by CPU idle state.

Thank you, this patchset looks fine to me and is a step in the right
direction:

Reviewed-by: Bartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com>

> Some Non-CPU device must need to monitor the device status like
> utilization regardless of CPU state.

This is probably true for all devfreq devices using simple_ondemand
governor by default:

drivers/devfreq/exynos-bus.c
drivers/devfreq/rk3399_dmc.c
drivers/devfreq/tegra20-devfreq.c
drivers/gpu/drm/lima/lima_devfreq.c
drivers/gpu/drm/msm/msm_gpu.c
drivers/gpu/drm/panfrost/panfrost_devfreq.c
drivers/memory/samsung/exynos5422-dmc.c
drivers/scsi/ufs/ufshcd.c

With devfreq device polling being "coupled" to CPU idle state
the devfreq subsystem behavior is completely unpredictable and
unreliable.

It affects both performance (device opp change up happening too
late) and power consumption (device opp change down happening too
late).

It also causes hardware usage counters support to report too high
values (because of CPU idle "coupling" the real polling period
becomes larger than maximum period supported by the counter and
the counter becomes fully "saturated") which negatively affects
power consumption (as has been observed when using Odroid XU3/4).

[ The only upside of using such "coupling" is lowered CPU power
  usage (in some situations) but at the (unacceptable IMHO) cost
  of the correctness of operations of devfreq subsystem. ]

Unfortunately this patchset currently fixes only exynos5422-dmc
devfreq driver. To fix problems for Exynos platforms we need to
also fix exynos-bus devfreq driver.

Best regards,
--
Bartlomiej Zolnierkiewicz
Samsung R&D Institute Poland
Samsung Electronics

> - patch1 explains the detailed reason why the delayed timer is required.
> - patch2 initializes that exynos5422-dmc device use delayed timer as default
> instead of deferrable timer.
> 
> Chanwoo Choi (2):
>   PM / devfreq: Add support delayed timer for polling mode
>   memory: samsung: exynos5422-dmc: Use delayed timer as default
> 
>  Documentation/ABI/testing/sysfs-class-devfreq | 12 +++
>  drivers/devfreq/devfreq.c                     | 83 ++++++++++++++++++-
>  drivers/memory/samsung/exynos5422-dmc.c       |  1 +
>  include/linux/devfreq.h                       |  9 ++
>  4 files changed, 104 insertions(+), 1 deletion(-)

^ permalink raw reply	[flat|nested] 11+ messages in thread

* Re: [RFC PATCH 2/2] memory: samsung: exynos5422-dmc: Use delayed timer as default
  2020-07-03  6:26     ` [RFC PATCH 2/2] memory: samsung: exynos5422-dmc: Use delayed timer as default Chanwoo Choi
@ 2020-07-08 12:33       ` Krzysztof Kozlowski
  2020-07-09  5:44         ` Chanwoo Choi
  0 siblings, 1 reply; 11+ messages in thread
From: Krzysztof Kozlowski @ 2020-07-08 12:33 UTC (permalink / raw)
  To: Chanwoo Choi
  Cc: lukasz.luba, k.konieczny, kgene, s.nawrocki, willy.mh.wolff.ml,
	b.zolnierkie, chanwoo, myungjoo.ham, kyungmin.park, linux-pm,
	linux-kernel, linux-samsung-soc, linux-arm-kernel

On Fri, Jul 03, 2020 at 03:26:22PM +0900, Chanwoo Choi wrote:
> Use delayed timer as default instead of deferrable timer
> in order to monitor the DMC status regardless of CPU idle.
> 
> Signed-off-by: Chanwoo Choi <cw00.choi@samsung.com>
> ---
>  drivers/memory/samsung/exynos5422-dmc.c | 1 +
>  1 file changed, 1 insertion(+)

If you want to take this through devfreq tree, I'm fine:
Reviewed-by: Krzysztof Kozlowski <krzk@kernel.org>

Otherwise I could grab it through samsung-soc after devfreq change ends
in mainline.

Best regards,
Krzysztof

^ permalink raw reply	[flat|nested] 11+ messages in thread

* Re: [RFC PATCH 0/2] PM / devfreq: Add delayed timer for polling
  2020-07-03  6:26 ` [RFC PATCH 0/2] PM / devfreq: Add delayed timer for polling Chanwoo Choi
                     ` (3 preceding siblings ...)
  2020-07-08 10:52   ` Bartlomiej Zolnierkiewicz
@ 2020-07-08 14:01   ` Lukasz Luba
  4 siblings, 0 replies; 11+ messages in thread
From: Lukasz Luba @ 2020-07-08 14:01 UTC (permalink / raw)
  To: Chanwoo Choi, k.konieczny, krzk, kgene
  Cc: s.nawrocki, willy.mh.wolff.ml, b.zolnierkie, chanwoo,
	myungjoo.ham, kyungmin.park, linux-pm, linux-kernel,
	linux-samsung-soc, linux-arm-kernel

Hi all,

On 7/3/20 7:26 AM, Chanwoo Choi wrote:
> Add the delayed timer to devfreq framework in order to support
> the periodical polling mode without stop caused by CPU idle state.
> Some Non-CPU device must need to monitor the device status like
> utilization regardless of CPU state.
> 
> - patch1 explains the detailed reason why the delayed timer is required.
> - patch2 initializes that exynos5422-dmc device use delayed timer as default
> instead of deferrable timer.
> 
> Chanwoo Choi (2):
>    PM / devfreq: Add support delayed timer for polling mode
>    memory: samsung: exynos5422-dmc: Use delayed timer as default
> 
>   Documentation/ABI/testing/sysfs-class-devfreq | 12 +++
>   drivers/devfreq/devfreq.c                     | 83 ++++++++++++++++++-
>   drivers/memory/samsung/exynos5422-dmc.c       |  1 +
>   include/linux/devfreq.h                       |  9 ++
>   4 files changed, 104 insertions(+), 1 deletion(-)
> 


My apologizes for being late for the party. I wasn't able to run tests
and I had to fix my setup after I messed up some scripts.

The patch set looks good to me, so you can add my (to both patches):

Reviewed-by: Lukasz Luba <lukasz.luba@arm.com>

I have run these Willy's benchmark tests and I will send some
follow up patches in a few minutes.

Regards,
Lukasz

^ permalink raw reply	[flat|nested] 11+ messages in thread

* Re: [RFC PATCH 0/2] PM / devfreq: Add delayed timer for polling
  2020-07-03 12:33   ` [RFC PATCH 0/2] PM / devfreq: Add delayed timer for polling Willy Wolff
@ 2020-07-08 14:25     ` Lukasz Luba
  2020-07-10 15:12       ` Willy Wolff
  0 siblings, 1 reply; 11+ messages in thread
From: Lukasz Luba @ 2020-07-08 14:25 UTC (permalink / raw)
  To: Willy Wolff, Chanwoo Choi
  Cc: k.konieczny, krzk, kgene, s.nawrocki, b.zolnierkie, chanwoo,
	myungjoo.ham, kyungmin.park, linux-pm, linux-kernel,
	linux-samsung-soc, linux-arm-kernel

Hi Willy,

On 7/3/20 1:33 PM, Willy Wolff wrote:
> Hi Chanwoo,
> 
> I think it doesn't help on the benchmark I suggested that is doing only memory
> accesses. With both timer, I have the same timing.
> 
> To test the benchmark with these new patches about timer:
> 
> git clone https://github.com/wwilly/benchmark.git \
>    && cd benchmark \
>    && source env.sh \
>    && ./bench_build.sh \
>    && bash source/scripts/test_dvfs_mem_patched.sh
> 
> The benchmark is set by default to run for 1s, but you can increase this by
> tweaking the script as:
> 
> taskset 8 ./bench_install/bin/microbe_cache 33554431 0 9722222 <TIME in sec> ${little_freq}
> 
> 
> Also, as I reported the issue, would it be possible to add a
> Reported-by: Willy Wolff <willy.mh.wolff.ml@gmail.com> ?
> Many thanks in advance.

Thank you for your good work and the benchmark. I hope you will continue
to use it and report some issues. I am going to send a follow up patches
for the DMC and I will add your 'Reported-by'. In the tests I can see
the improvements, but it's worth to consult with you if I understand
the new results correctly.

I think there is still some area for improvements in the devfreq and you
could find the interesting bits to contribute.

Regards,
Lukasz

> 
> 
> Best Regards,
> Willy
> 

^ permalink raw reply	[flat|nested] 11+ messages in thread

* Re: [RFC PATCH 2/2] memory: samsung: exynos5422-dmc: Use delayed timer as default
  2020-07-08 12:33       ` Krzysztof Kozlowski
@ 2020-07-09  5:44         ` Chanwoo Choi
  0 siblings, 0 replies; 11+ messages in thread
From: Chanwoo Choi @ 2020-07-09  5:44 UTC (permalink / raw)
  To: Krzysztof Kozlowski
  Cc: lukasz.luba, k.konieczny, kgene, s.nawrocki, willy.mh.wolff.ml,
	b.zolnierkie, chanwoo, myungjoo.ham, kyungmin.park, linux-pm,
	linux-kernel, linux-samsung-soc, linux-arm-kernel

Hi Krzysztof,

On 7/8/20 9:33 PM, Krzysztof Kozlowski wrote:
> On Fri, Jul 03, 2020 at 03:26:22PM +0900, Chanwoo Choi wrote:
>> Use delayed timer as default instead of deferrable timer
>> in order to monitor the DMC status regardless of CPU idle.
>>
>> Signed-off-by: Chanwoo Choi <cw00.choi@samsung.com>
>> ---
>>  drivers/memory/samsung/exynos5422-dmc.c | 1 +
>>  1 file changed, 1 insertion(+)
> 
> If you want to take this through devfreq tree, I'm fine:
> Reviewed-by: Krzysztof Kozlowski <krzk@kernel.org>

OK. I'll take it to devfreq-next for v5.9-rc1. Thanks.

> 
> Otherwise I could grab it through samsung-soc after devfreq change ends
> in mainline.
> 
> Best regards,
> Krzysztof
> 
> 


-- 
Best Regards,
Chanwoo Choi
Samsung Electronics

^ permalink raw reply	[flat|nested] 11+ messages in thread

* Re: [RFC PATCH 0/2] PM / devfreq: Add delayed timer for polling
  2020-07-08 14:25     ` Lukasz Luba
@ 2020-07-10 15:12       ` Willy Wolff
  2020-07-13  8:55         ` Lukasz Luba
  0 siblings, 1 reply; 11+ messages in thread
From: Willy Wolff @ 2020-07-10 15:12 UTC (permalink / raw)
  To: Lukasz Luba
  Cc: Chanwoo Choi, k.konieczny, krzk, kgene, s.nawrocki, b.zolnierkie,
	chanwoo, myungjoo.ham, kyungmin.park, linux-pm, linux-kernel,
	linux-samsung-soc, linux-arm-kernel

Hi Lukasz,

On 2020-07-08-15-25-03, Lukasz Luba wrote:
> Hi Willy,
> 
> On 7/3/20 1:33 PM, Willy Wolff wrote:
> > Hi Chanwoo,
> > 
> > I think it doesn't help on the benchmark I suggested that is doing only memory
> > accesses. With both timer, I have the same timing.
> > 
> > To test the benchmark with these new patches about timer:
> > 
> > git clone https://github.com/wwilly/benchmark.git \
> >    && cd benchmark \
> >    && source env.sh \
> >    && ./bench_build.sh \
> >    && bash source/scripts/test_dvfs_mem_patched.sh
> > 
> > The benchmark is set by default to run for 1s, but you can increase this by
> > tweaking the script as:
> > 
> > taskset 8 ./bench_install/bin/microbe_cache 33554431 0 9722222 <TIME in sec> ${little_freq}
> > 
> > 
> > Also, as I reported the issue, would it be possible to add a
> > Reported-by: Willy Wolff <willy.mh.wolff.ml@gmail.com> ?
> > Many thanks in advance.
> 
> Thank you for your good work and the benchmark. I hope you will continue
> to use it and report some issues. I am going to send a follow up patches
> for the DMC and I will add your 'Reported-by'. In the tests I can see
> the improvements, but it's worth to consult with you if I understand
> the new results correctly.
> 

Thanks for that. I will follow on the other patch thread discussion.

> I think there is still some area for improvements in the devfreq and you
> could find the interesting bits to contribute.

In fact, this benchmark is motivated about part of my PhD research that has just
been accepted at LCTES2020: "Performance Optimization on big.LITTLE Architectures:
A Memory-latency Aware Approach" at https://dl.acm.org/doi/10.1145/3372799.3394370

Basically, it's about snooping latency with "bad" CPU DVFS choice on big.LITTLE
systems or more generally SMP/AMP architecture. I'm cleaning up my code and will
propose patches as an RFC later. It introduces a new CPU DVFS governor to limit
snooping latency.

Cheers,
Willy

> 
> Regards,
> Lukasz
> 
> > 
> > 
> > Best Regards,
> > Willy
> > 

^ permalink raw reply	[flat|nested] 11+ messages in thread

* Re: [RFC PATCH 0/2] PM / devfreq: Add delayed timer for polling
  2020-07-10 15:12       ` Willy Wolff
@ 2020-07-13  8:55         ` Lukasz Luba
  0 siblings, 0 replies; 11+ messages in thread
From: Lukasz Luba @ 2020-07-13  8:55 UTC (permalink / raw)
  To: Willy Wolff
  Cc: Chanwoo Choi, k.konieczny, krzk, kgene, s.nawrocki, b.zolnierkie,
	chanwoo, myungjoo.ham, kyungmin.park, linux-pm, linux-kernel,
	linux-samsung-soc, linux-arm-kernel

Hi Willy

On 7/10/20 4:12 PM, Willy Wolff wrote:
> Hi Lukasz,
> 
> On 2020-07-08-15-25-03, Lukasz Luba wrote:
>> Hi Willy,
>>
>> On 7/3/20 1:33 PM, Willy Wolff wrote:
>>> Hi Chanwoo,
>>>
>>> I think it doesn't help on the benchmark I suggested that is doing only memory
>>> accesses. With both timer, I have the same timing.
>>>
>>> To test the benchmark with these new patches about timer:
>>>
>>> git clone https://github.com/wwilly/benchmark.git \
>>>     && cd benchmark \
>>>     && source env.sh \
>>>     && ./bench_build.sh \
>>>     && bash source/scripts/test_dvfs_mem_patched.sh
>>>
>>> The benchmark is set by default to run for 1s, but you can increase this by
>>> tweaking the script as:
>>>
>>> taskset 8 ./bench_install/bin/microbe_cache 33554431 0 9722222 <TIME in sec> ${little_freq}
>>>
>>>
>>> Also, as I reported the issue, would it be possible to add a
>>> Reported-by: Willy Wolff <willy.mh.wolff.ml@gmail.com> ?
>>> Many thanks in advance.
>>
>> Thank you for your good work and the benchmark. I hope you will continue
>> to use it and report some issues. I am going to send a follow up patches
>> for the DMC and I will add your 'Reported-by'. In the tests I can see
>> the improvements, but it's worth to consult with you if I understand
>> the new results correctly.
>>
> 
> Thanks for that. I will follow on the other patch thread discussion.
> 
>> I think there is still some area for improvements in the devfreq and you
>> could find the interesting bits to contribute.
> 
> In fact, this benchmark is motivated about part of my PhD research that has just
> been accepted at LCTES2020: "Performance Optimization on big.LITTLE Architectures:
> A Memory-latency Aware Approach" at https://dl.acm.org/doi/10.1145/3372799.3394370
> 

Congrats and thank you for the link (I will read it).

> Basically, it's about snooping latency with "bad" CPU DVFS choice on big.LITTLE
> systems or more generally SMP/AMP architecture. I'm cleaning up my code and will
> propose patches as an RFC later. It introduces a new CPU DVFS governor to limit
> snooping latency.

This is interesting, please add me on CC in the patch set.

Regards,
Lukasz

^ permalink raw reply	[flat|nested] 11+ messages in thread

end of thread, back to index

Thread overview: 11+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
     [not found] <CGME20200703061508epcas1p171aa3c0ab832b77e5837d8bd1e563742@epcas1p1.samsung.com>
2020-07-03  6:26 ` [RFC PATCH 0/2] PM / devfreq: Add delayed timer for polling Chanwoo Choi
     [not found]   ` <CGME20200703061509epcas1p10446471df6497b4b2d7066280daa7f87@epcas1p1.samsung.com>
2020-07-03  6:26     ` [RFC PATCH 1/2] PM / devfreq: Add support delayed timer for polling mode Chanwoo Choi
     [not found]   ` <CGME20200703061509epcas1p3e338b614c44f5d8b566dae92f11cc8cb@epcas1p3.samsung.com>
2020-07-03  6:26     ` [RFC PATCH 2/2] memory: samsung: exynos5422-dmc: Use delayed timer as default Chanwoo Choi
2020-07-08 12:33       ` Krzysztof Kozlowski
2020-07-09  5:44         ` Chanwoo Choi
2020-07-03 12:33   ` [RFC PATCH 0/2] PM / devfreq: Add delayed timer for polling Willy Wolff
2020-07-08 14:25     ` Lukasz Luba
2020-07-10 15:12       ` Willy Wolff
2020-07-13  8:55         ` Lukasz Luba
2020-07-08 10:52   ` Bartlomiej Zolnierkiewicz
2020-07-08 14:01   ` Lukasz Luba

Linux-Samsung-soc Archive on lore.kernel.org

Archives are clonable:
	git clone --mirror https://lore.kernel.org/linux-samsung-soc/0 linux-samsung-soc/git/0.git

	# If you have public-inbox 1.1+ installed, you may
	# initialize and index your mirror using the following commands:
	public-inbox-init -V2 linux-samsung-soc linux-samsung-soc/ https://lore.kernel.org/linux-samsung-soc \
		linux-samsung-soc@vger.kernel.org
	public-inbox-index linux-samsung-soc

Example config snippet for mirrors

Newsgroup available over NNTP:
	nntp://nntp.lore.kernel.org/org.kernel.vger.linux-samsung-soc


AGPL code for this site: git clone https://public-inbox.org/public-inbox.git