[PATCH v3 0/4] thermal: re-calculate k_po/k_pu when update sustainable power

[PATCH v3 0/4] thermal: re-calculate k_po/k_pu when update sustainable power

[Leo Yan]

k_po/k_pu are two proportional term constants and essentially they have
fixed ratio compared with sustainable power. In current implementation,
k_po and k_pu are absolute value after calculation and cannot represent
the ratio relationship with sustainable power; as a result, when change
sustainable power we cannot smoothly change proportional term constant.

So this patch series introduces k_po_ratio and k_pu_ratio, which
represent the ratio value compared against sustainable power. Also add
sys file system nodes for them for easily update them from userspace and
update a bit in documentation.

Changes from v2:
* According to Eduardo' comments, move code from thermal_core to
  power_allocator file
* According to Daniel's review, v1 will introduce accumulated rounding
  errors; v2 patches can dismiss this issue

Changes from v1:
* Fix compiling error

Leo Yan (4):
  thermal: power_allocator: rework proportional parameter
  thermal: power_allocator: change k_pu_ratio/k_po_ratio as percentage
  thermal: add sys node for k_pu_ratio/k_po_ratio
  thermal: power_allocator: document k_pu_ratio/k_po_ratio

 Documentation/thermal/power_allocator.txt | 15 +++++++++++----
 drivers/thermal/power_allocator.c         | 18 ++++++++++++------
 drivers/thermal/thermal_core.c            | 23 +++++++++++++++++++++--
 include/linux/thermal.h                   |  4 ++--
 4 files changed, 46 insertions(+), 14 deletions(-)

-- 
1.9.1

[PATCH v3 1/4] thermal: power_allocator: rework proportional parameter

[Leo Yan]

k_po/k_pu are two proportional term constants and essentially they have
fixed ratio compared with sustainable power. In current implementation,
k_po and k_pu are absolute value after calculation and cannot represent
the ratio relationship with sustainable power; as a result, when change
sustainable power we cannot smoothly change proportional term constant.

So this patch introduces variants k_po_ratio and k_pu_ratio, which
represent the ratio value compared against sustainable power, so smaller
ratio will result in thermal slower ramp. The formulas finally are as
below:

k_pu = k_pu_ratio * sustainable_power /
		(desired_temperature - switch_on_temp)

k_po = k_po_ratio * sustainable_power /
		(desired_temperature - switch_on_temp)

Signed-off-by: Leo Yan <leo.yan@linaro.org>
---
 drivers/thermal/power_allocator.c | 18 ++++++++++++------
 include/linux/thermal.h           |  4 ++--
 2 files changed, 14 insertions(+), 8 deletions(-)

diff --git a/drivers/thermal/power_allocator.c b/drivers/thermal/power_allocator.c
index 1246aa6..76d2b47 100644
--- a/drivers/thermal/power_allocator.c
+++ b/drivers/thermal/power_allocator.c
@@ -155,13 +155,11 @@ static void estimate_pid_constants(struct thermal_zone_device *tz,
 	if (!temperature_threshold)
 		return;
 
-	if (!tz->tzp->k_po || force)
-		tz->tzp->k_po = int_to_frac(sustainable_power) /
-			temperature_threshold;
+	if (!tz->tzp->k_po_ratio || force)
+		tz->tzp->k_po_ratio = 1;
 
-	if (!tz->tzp->k_pu || force)
-		tz->tzp->k_pu = int_to_frac(2 * sustainable_power) /
-			temperature_threshold;
+	if (!tz->tzp->k_pu_ratio || force)
+		tz->tzp->k_pu_ratio = 2;
 
 	if (!tz->tzp->k_i || force)
 		tz->tzp->k_i = int_to_frac(10) / 1000;
@@ -169,6 +167,11 @@ static void estimate_pid_constants(struct thermal_zone_device *tz,
 	 * The default for k_d and integral_cutoff is 0, so we can
 	 * leave them as they are.
 	 */
+
+	tz->tzp->k_po = int_to_frac(tz->tzp->k_po_ratio * sustainable_power) /
+				temperature_threshold;
+	tz->tzp->k_pu = int_to_frac(tz->tzp->k_pu_ratio * sustainable_power) /
+				temperature_threshold;
 }
 
 /**
@@ -202,6 +205,9 @@ static u32 pid_controller(struct thermal_zone_device *tz,
 
 	if (tz->tzp->sustainable_power) {
 		sustainable_power = tz->tzp->sustainable_power;
+		estimate_pid_constants(tz, sustainable_power,
+				       params->trip_switch_on, control_temp,
+				       false);
 	} else {
 		sustainable_power = estimate_sustainable_power(tz);
 		estimate_pid_constants(tz, sustainable_power,
diff --git a/include/linux/thermal.h b/include/linux/thermal.h
index 613c29b..2638f0b 100644
--- a/include/linux/thermal.h
+++ b/include/linux/thermal.h
@@ -284,13 +284,13 @@ struct thermal_zone_params {
 	 * Proportional parameter of the PID controller when
 	 * overshooting (i.e., when temperature is below the target)
 	 */
-	s32 k_po;
+	s32 k_po, k_po_ratio;
 
 	/*
 	 * Proportional parameter of the PID controller when
 	 * undershooting
 	 */
-	s32 k_pu;
+	s32 k_pu, k_pu_ratio;
 
 	/* Integral parameter of the PID controller */
 	s32 k_i;
-- 
1.9.1

[PATCH v3 2/4] thermal: power_allocator: change k_pu_ratio/k_po_ratio as percentage

[Leo Yan]

Change k_pu_ratio/k_po_ratio as percentage value, so can set these
values from sys file nodes with high resolution.

Signed-off-by: Leo Yan <leo.yan@linaro.org>
---
 drivers/thermal/power_allocator.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/drivers/thermal/power_allocator.c b/drivers/thermal/power_allocator.c
index 76d2b47..c5f36fa 100644
--- a/drivers/thermal/power_allocator.c
+++ b/drivers/thermal/power_allocator.c
@@ -156,10 +156,10 @@ static void estimate_pid_constants(struct thermal_zone_device *tz,
 		return;
 
 	if (!tz->tzp->k_po_ratio || force)
-		tz->tzp->k_po_ratio = 1;
+		tz->tzp->k_po_ratio = 100;
 
 	if (!tz->tzp->k_pu_ratio || force)
-		tz->tzp->k_pu_ratio = 2;
+		tz->tzp->k_pu_ratio = 200;
 
 	if (!tz->tzp->k_i || force)
 		tz->tzp->k_i = int_to_frac(10) / 1000;
@@ -169,9 +169,9 @@ static void estimate_pid_constants(struct thermal_zone_device *tz,
 	 */
 
 	tz->tzp->k_po = int_to_frac(tz->tzp->k_po_ratio * sustainable_power) /
-				temperature_threshold;
+				temperature_threshold / 100;
 	tz->tzp->k_pu = int_to_frac(tz->tzp->k_pu_ratio * sustainable_power) /
-				temperature_threshold;
+				temperature_threshold / 100;
 }
 
 /**
-- 
1.9.1

[PATCH v3 3/4] thermal: add sys node for k_pu_ratio/k_po_ratio

[Leo Yan]

Add two sys nodes for k_pu_ratio and k_po_ratio, so user can set ratio
value from them. Also change k_pu/k_po as read only nodes, it will only
be used to show proportional term constants which are calculated by
ratio.

Signed-off-by: Leo Yan <leo.yan@linaro.org>
---
 drivers/thermal/thermal_core.c | 23 +++++++++++++++++++++--
 1 file changed, 21 insertions(+), 2 deletions(-)

diff --git a/drivers/thermal/thermal_core.c b/drivers/thermal/thermal_core.c
index d9e525c..bc149a67 100644
--- a/drivers/thermal/thermal_core.c
+++ b/drivers/thermal/thermal_core.c
@@ -955,8 +955,25 @@ static DEVICE_ATTR(sustainable_power, S_IWUSR | S_IRUGO, sustainable_power_show,
 	}								\
 	static DEVICE_ATTR(name, S_IWUSR | S_IRUGO, name##_show, name##_store)
 
-create_s32_tzp_attr(k_po);
-create_s32_tzp_attr(k_pu);
+
+#define create_s32_tzp_attr_ro(name)					\
+	static ssize_t							\
+	name##_show(struct device *dev, struct device_attribute *devattr, \
+		char *buf)						\
+	{								\
+	struct thermal_zone_device *tz = to_thermal_zone(dev);		\
+									\
+	if (tz->tzp)							\
+		return sprintf(buf, "%u\n", tz->tzp->name);		\
+	else								\
+		return -EIO;						\
+	}								\
+	static DEVICE_ATTR(name, S_IRUGO, name##_show, NULL)
+
+create_s32_tzp_attr_ro(k_po);
+create_s32_tzp_attr_ro(k_pu);
+create_s32_tzp_attr(k_po_ratio);
+create_s32_tzp_attr(k_pu_ratio);
 create_s32_tzp_attr(k_i);
 create_s32_tzp_attr(k_d);
 create_s32_tzp_attr(integral_cutoff);
@@ -968,6 +985,8 @@ static struct device_attribute *dev_tzp_attrs[] = {
 	&dev_attr_sustainable_power,
 	&dev_attr_k_po,
 	&dev_attr_k_pu,
+	&dev_attr_k_po_ratio,
+	&dev_attr_k_pu_ratio,
 	&dev_attr_k_i,
 	&dev_attr_k_d,
 	&dev_attr_integral_cutoff,
-- 
1.9.1

[PATCH v3 4/4] thermal: power_allocator: document k_pu_ratio/k_po_ratio

[Leo Yan]

Document k_pu_ratio and k_po_ratio's purpose, and refine formula to
demonstrate to use ratio to calculate k_pu/k_po.

Signed-off-by: Leo Yan <leo.yan@linaro.org>
---
 Documentation/thermal/power_allocator.txt | 15 +++++++++++----
 1 file changed, 11 insertions(+), 4 deletions(-)

diff --git a/Documentation/thermal/power_allocator.txt b/Documentation/thermal/power_allocator.txt
index a1ce223..8958dab 100644
--- a/Documentation/thermal/power_allocator.txt
+++ b/Documentation/thermal/power_allocator.txt
@@ -114,13 +114,20 @@ whilst temperature is low, and may lead to temperature overshooting.
 
 The default value for `k_pu` is:
 
-    2 * sustainable_power / (desired_temperature - switch_on_temp)
+    k_pu_ratio * sustainable_power / (desired_temperature - switch_on_temp)
 
 This means that at `switch_on_temp` the output of the controller's
-proportional term will be 2 * `sustainable_power`.  The default value
-for `k_po` is:
+proportional term will be k_pu_ratio * `sustainable_power`.
 
-    sustainable_power / (desired_temperature - switch_on_temp)
+The default value for `k_po` is:
+
+    k_po_ratio * sustainable_power / (desired_temperature - switch_on_temp)
+
+k_pu_ratio's default value is 2 and k_po_ratio's default value is 1.
+These two vairants mean the ratio value compared against sustainable power,
+so smaller ratio will result in thermal slower ramp. User can set these two
+ratio value from sys filesystem nodes, finally k_pu and k_po will be
+calculated by updated ratio value.
 
 Focusing on the proportional and feed forward values of the PID
 controller equation we have:
-- 
1.9.1

Re: [PATCH v3 0/4] thermal: re-calculate k_po/k_pu when update sustainable power

[Javi Merino]

On Wed, Jan 06, 2016 at 04:53:44PM +0800, Leo Yan wrote:
> k_po/k_pu are two proportional term constants and essentially they have
> fixed ratio compared with sustainable power. In current implementation,
> k_po and k_pu are absolute value after calculation and cannot represent
> the ratio relationship with sustainable power; as a result, when change
> sustainable power we cannot smoothly change proportional term constant.

In v2 Daniel said that the use case was made up.  Can you elaborate on
why we need this?

> So this patch series introduces k_po_ratio and k_pu_ratio, which
> represent the ratio value compared against sustainable power. Also add
> sys file system nodes for them for easily update them from userspace and
> update a bit in documentation.

Actually, it makes it harder to update from userspace.  Now userspace
can't set a k_po/k_pu any more, and it's forced to set them as ratios
of sustainable power. I'd rather not do this unless there is a good
reason for it.

Cheers,
Javi

Re: [PATCH v3 0/4] thermal: re-calculate k_po/k_pu when update sustainable power

[Leo Yan]

Hi Javi,

On Wed, Jan 06, 2016 at 10:07:13AM +0000, Javi Merino wrote:
> On Wed, Jan 06, 2016 at 04:53:44PM +0800, Leo Yan wrote:
> > k_po/k_pu are two proportional term constants and essentially they have
> > fixed ratio compared with sustainable power. In current implementation,
> > k_po and k_pu are absolute value after calculation and cannot represent
> > the ratio relationship with sustainable power; as a result, when change
> > sustainable power we cannot smoothly change proportional term constant.
> 
> In v2 Daniel said that the use case was made up.  Can you elaborate on
> why we need this?

When I did IPA profiling on Hikey, I tried to set different
sustainable power. But after changed sustainable power, k_po and k_pu
were stale value which calculated by old sustainable power.

So with this patch, it's more convinence for profiling and can avoid
confusion for k_po/k_pu after update sustainable power.

> > So this patch series introduces k_po_ratio and k_pu_ratio, which
> > represent the ratio value compared against sustainable power. Also add
> > sys file system nodes for them for easily update them from userspace and
> > update a bit in documentation.
> 
> Actually, it makes it harder to update from userspace.  Now userspace
> can't set a k_po/k_pu any more, and it's forced to set them as ratios
> of sustainable power. I'd rather not do this unless there is a good
> reason for it.

We use ratios of sustainable power to calculate k_po/k_pu, so if user
explicitly specify ratio values we can easily to update k_po/k_pu
after we change sustainable power.

Otherwise, it's hard to get to know the real ratio values of
sustainable power. Please review below case:

- If we set new sustainable power, then actually in kernel it should
  update k_po/k_pu so that can keep fixed ration;
- At the meantime, user can set k_po/k_pu from sys file node so that
  eventually change ratio;

So here has race condition with updating k_po/k_pu and the code is
hard to distinguish what is real real ratio. If have explicit parameters
for ratios, then will be much clear.

Thanks,
Leo Yan

Re: [PATCH v3 0/4] thermal: re-calculate k_po/k_pu when update sustainable power

[Daniel Thompson]

On 06/01/16 10:07, Javi Merino wrote:
> On Wed, Jan 06, 2016 at 04:53:44PM +0800, Leo Yan wrote:
>> k_po/k_pu are two proportional term constants and essentially they have
>> fixed ratio compared with sustainable power. In current implementation,
>> k_po and k_pu are absolute value after calculation and cannot represent
>> the ratio relationship with sustainable power; as a result, when change
>> sustainable power we cannot smoothly change proportional term constant.
>
> In v2 Daniel said that the use case was made up.  Can you elaborate on
> why we need this?

To be clear, I didn't say Leo's use case was made up. In truth I've 
never asked Leo what his use case is.

It was simply the use case that I used to illustrate a bug in the v1 
implementation which was made up (and which you asked for more details of).



Daniel.

Re: [PATCH v3 0/4] thermal: re-calculate k_po/k_pu when update sustainable power

[Javi Merino]

On Wed, Jan 06, 2016 at 11:21:54AM +0000, Daniel Thompson wrote:
> On 06/01/16 10:07, Javi Merino wrote:
> >On Wed, Jan 06, 2016 at 04:53:44PM +0800, Leo Yan wrote:
> >>k_po/k_pu are two proportional term constants and essentially they have
> >>fixed ratio compared with sustainable power. In current implementation,
> >>k_po and k_pu are absolute value after calculation and cannot represent
> >>the ratio relationship with sustainable power; as a result, when change
> >>sustainable power we cannot smoothly change proportional term constant.
> >
> >In v2 Daniel said that the use case was made up.  Can you elaborate on
> >why we need this?
> 
> To be clear, I didn't say Leo's use case was made up. In truth I've
> never asked Leo what his use case is.

Sorry, I didn't mean to say that Leo had made it up.  I was just
asking for a use case and it looked to me that the "fan inhibitor
mode" was something that you guys had talked about.

Cheers,
Javi

> It was simply the use case that I used to illustrate a bug in the v1
> implementation which was made up (and which you asked for more
> details of).
> 
> 
> 
> Daniel.
>