cpufreq on-demand governor up_treshold?

cpufreq on-demand governor up_treshold?

[Jan De Luyck]

Hello lists,

(please cc me from cpufreq list)

I've since yesterday started using the ondemand governor. Seems to work fine, 
tho I can't seem to find a reason why it keeps scaling my processor speed 
upwards tho the processor use never exceeds 30% (been watching top -d 1). 

I've been using the powernowd daemon and the userspace governor previously, 
which doesn't seem to have this problem, even if i set it at the same 
sampling rates as the ondemand governor.

The settings in /sys/.../ondemand are default.

Any hints?

Thanks.
-- 
Women aren't as mere as they used to be.
  -- Pogo

Re: cpufreq on-demand governor up_treshold?

[Eric Piel]

[-- Attachment #1: Type: text/plain, Size: 1065 bytes --]

Jan De Luyck a écrit :
> Hello lists,
> 
> (please cc me from cpufreq list)
> 
> I've since yesterday started using the ondemand governor. Seems to work fine, 
> tho I can't seem to find a reason why it keeps scaling my processor speed 
> upwards tho the processor use never exceeds 30% (been watching top -d 1). 
:
:
> Any hints?
You can try the three attached patches in the order :
ondemand-cleanup-factorise-idle-measurement-2.6.11.patch
ondemand-save-idle-up-for-all-cpu-2.6.11.patch
ondemand-automatic-downscaling-2.6.11-accepted.patch

They are available on the cpufreq list but as it's difficult to access 
it I'm sending them again, all together. These are the last things that 
Venki and I have been working on. It should solve your problem 
(actually, only the last patch, but it depends on the two previous 
patches). Please, let me know if it works.

BTW, DaveJ, Dominik, I couldn't find them in the daily-snapshot 
available at codemonkey.org.uk. Should I worry, or is it just due to 
some latency between your private trees and the public one?

Eric

[-- Attachment #2: ondemand-automatic-downscaling-2.6.11-accepted.patch --]
[-- Type: text/x-patch, Size: 5003 bytes --]

diff -purN linux-2.6.11/drivers/cpufreq/cpufreq_ondemand.c linux-2.6.11-new/drivers/cpufreq/cpufreq_ondemand.c
--- linux-2.6.11/drivers/cpufreq/cpufreq_ondemand.c	2005-03-07 17:57:31.000000000 -0800
+++ linux-2.6.11-new/drivers/cpufreq/cpufreq_ondemand.c	2005-03-07 17:53:17.000000000 -0800
@@ -34,13 +34,9 @@
  */
 
 #define DEF_FREQUENCY_UP_THRESHOLD		(80)
-#define MIN_FREQUENCY_UP_THRESHOLD		(0)
+#define MIN_FREQUENCY_UP_THRESHOLD		(10)
 #define MAX_FREQUENCY_UP_THRESHOLD		(100)
 
-#define DEF_FREQUENCY_DOWN_THRESHOLD		(20)
-#define MIN_FREQUENCY_DOWN_THRESHOLD		(0)
-#define MAX_FREQUENCY_DOWN_THRESHOLD		(100)
-
 /* 
  * The polling frequency of this governor depends on the capability of 
  * the processor. Default polling frequency is 1000 times the transition
@@ -78,12 +74,10 @@ struct dbs_tuners {
 	unsigned int 		sampling_rate;
 	unsigned int		sampling_down_factor;
 	unsigned int		up_threshold;
-	unsigned int		down_threshold;
 };
 
 static struct dbs_tuners dbs_tuners_ins = {
 	.up_threshold 		= DEF_FREQUENCY_UP_THRESHOLD,
-	.down_threshold 	= DEF_FREQUENCY_DOWN_THRESHOLD,
 	.sampling_down_factor 	= DEF_SAMPLING_DOWN_FACTOR,
 };
 
@@ -115,7 +109,6 @@ static ssize_t show_##file_name						\
 show_one(sampling_rate, sampling_rate);
 show_one(sampling_down_factor, sampling_down_factor);
 show_one(up_threshold, up_threshold);
-show_one(down_threshold, down_threshold);
 
 static ssize_t store_sampling_down_factor(struct cpufreq_policy *unused, 
 		const char *buf, size_t count)
@@ -161,8 +154,7 @@ static ssize_t store_up_threshold(struct
 
 	down(&dbs_sem);
 	if (ret != 1 || input > MAX_FREQUENCY_UP_THRESHOLD || 
-			input < MIN_FREQUENCY_UP_THRESHOLD ||
-			input <= dbs_tuners_ins.down_threshold) {
+			input < MIN_FREQUENCY_UP_THRESHOLD) {
 		up(&dbs_sem);
 		return -EINVAL;
 	}
@@ -173,26 +165,6 @@ static ssize_t store_up_threshold(struct
 	return count;
 }
 
-static ssize_t store_down_threshold(struct cpufreq_policy *unused, 
-		const char *buf, size_t count)
-{
-	unsigned int input;
-	int ret;
-	ret = sscanf (buf, "%u", &input);
-
-	down(&dbs_sem);
-	if (ret != 1 || input > MAX_FREQUENCY_DOWN_THRESHOLD || 
-			input < MIN_FREQUENCY_DOWN_THRESHOLD ||
-			input >= dbs_tuners_ins.up_threshold) {
-		up(&dbs_sem);
-		return -EINVAL;
-	}
-
-	dbs_tuners_ins.down_threshold = input;
-	up(&dbs_sem);
-
-	return count;
-}
 
 #define define_one_rw(_name) \
 static struct freq_attr _name = \
@@ -201,7 +173,6 @@ __ATTR(_name, 0644, show_##_name, store_
 define_one_rw(sampling_rate);
 define_one_rw(sampling_down_factor);
 define_one_rw(up_threshold);
-define_one_rw(down_threshold);
 
 static struct attribute * dbs_attributes[] = {
 	&sampling_rate_max.attr,
@@ -209,7 +180,6 @@ static struct attribute * dbs_attributes
 	&sampling_rate.attr,
 	&sampling_down_factor.attr,
 	&up_threshold.attr,
-	&down_threshold.attr,
 	NULL
 };
 
@@ -222,8 +192,8 @@ static struct attribute_group dbs_attr_g
 
 static void dbs_check_cpu(int cpu)
 {
-	unsigned int idle_ticks, up_idle_ticks, down_idle_ticks;
-	unsigned int freq_down_step;
+	unsigned int idle_ticks, up_idle_ticks, total_ticks;
+	unsigned int freq_next;
 	unsigned int freq_down_sampling_rate;
 	static int down_skip[NR_CPUS];
 	struct cpu_dbs_info_s *this_dbs_info;
@@ -290,7 +260,12 @@ static void dbs_check_cpu(int cpu)
 	down_skip[cpu]++;
 	if (down_skip[cpu] < dbs_tuners_ins.sampling_down_factor)
 		return;
+	down_skip[cpu] = 0;
 
+	/* don't try to decrease the frequency if it's already the min */
+	if (policy->cur == policy->min)
+		return;
+	
 	idle_ticks = UINT_MAX;
 	for_each_cpu_mask(j, policy->cpus) {
 		unsigned int tmp_idle_ticks, total_idle_ticks;
@@ -308,27 +283,23 @@ static void dbs_check_cpu(int cpu)
 			idle_ticks = tmp_idle_ticks;
 	}
 
-	/* Scale idle ticks by 100 and compare with up and down ticks */
-	idle_ticks *= 100;
-	down_skip[cpu] = 0;
-
+	/* Compute how many ticks there are between two measurements */
 	freq_down_sampling_rate = dbs_tuners_ins.sampling_rate *
 		dbs_tuners_ins.sampling_down_factor;
-	down_idle_ticks = (100 - dbs_tuners_ins.down_threshold) *
-			sampling_rate_in_HZ(freq_down_sampling_rate);
-
-	if (idle_ticks > down_idle_ticks) {
-		freq_down_step = (5 * policy->max) / 100;
-
-		/* max freq cannot be less than 100. But who knows.... */
-		if (unlikely(freq_down_step == 0))
-			freq_down_step = 5;
+	total_ticks = sampling_rate_in_HZ(freq_down_sampling_rate);
+	
+	/* 
+	 * The optimal frequency is the frequency that is the lowest that
+	 * can support the current CPU usage without triggering 
+	 * the up policy. To be safe, we focus 10 points under the threshold.
+	 */
+	freq_next = ((total_ticks - idle_ticks) * 100) / total_ticks;
+	freq_next = freq_next * policy->cur / (dbs_tuners_ins.up_threshold-10);
 
+	if (freq_next <= ((policy->cur * 95) / 100))
 		__cpufreq_driver_target(policy,
-			policy->cur - freq_down_step, 
-			CPUFREQ_RELATION_H);
-		return;
-	}
+			freq_next, 
+			CPUFREQ_RELATION_L);
 }
 
 static void do_dbs_timer(void *data)

[-- Attachment #3: ondemand-cleanup-factorise-idle-measurement-2.6.11.patch --]
[-- Type: text/x-patch, Size: 2487 bytes --]

--- linux-2.6.11/drivers/cpufreq/cpufreq_ondemand.c.bak	2005-02-06 23:35:41.000000000 +0100
+++ linux-2.6.11/drivers/cpufreq/cpufreq_ondemand.c	2005-03-06 19:04:12.000000000 +0100
@@ -223,7 +223,6 @@ static struct attribute_group dbs_attr_g
 static void dbs_check_cpu(int cpu)
 {
 	unsigned int idle_ticks, up_idle_ticks, down_idle_ticks;
-	unsigned int total_idle_ticks;
 	unsigned int freq_down_step;
 	unsigned int freq_down_sampling_rate;
 	static int down_skip[NR_CPUS];
@@ -252,20 +251,11 @@ static void dbs_check_cpu(int cpu)
 	 */
 
 	/* Check for frequency increase */
-	total_idle_ticks = kstat_cpu(cpu).cpustat.idle +
-		kstat_cpu(cpu).cpustat.iowait;
-	idle_ticks = total_idle_ticks -
-		this_dbs_info->prev_cpu_idle_up;
-	this_dbs_info->prev_cpu_idle_up = total_idle_ticks;
-	
-
+	idle_ticks = UINT_MAX;
 	for_each_cpu_mask(j, policy->cpus) {
-		unsigned int tmp_idle_ticks;
+		unsigned int tmp_idle_ticks, total_idle_ticks;
 		struct cpu_dbs_info_s *j_dbs_info;
 
-		if (j == cpu)
-			continue;
-
 		j_dbs_info = &per_cpu(cpu_dbs_info, j);
 		/* Check for frequency increase */
 		total_idle_ticks = kstat_cpu(j).cpustat.idle +
@@ -287,7 +277,7 @@ static void dbs_check_cpu(int cpu)
 		__cpufreq_driver_target(policy, policy->max, 
 			CPUFREQ_RELATION_H);
 		down_skip[cpu] = 0;
-		this_dbs_info->prev_cpu_idle_down = total_idle_ticks;
+		this_dbs_info->prev_cpu_idle_down = this_dbs_info->prev_cpu_idle_up;
 		return;
 	}
 
@@ -296,19 +286,11 @@ static void dbs_check_cpu(int cpu)
 	if (down_skip[cpu] < dbs_tuners_ins.sampling_down_factor)
 		return;
 
-	total_idle_ticks = kstat_cpu(cpu).cpustat.idle +
-		kstat_cpu(cpu).cpustat.iowait;
-	idle_ticks = total_idle_ticks -
-		this_dbs_info->prev_cpu_idle_down;
-	this_dbs_info->prev_cpu_idle_down = total_idle_ticks;
-
+	idle_ticks = UINT_MAX;
 	for_each_cpu_mask(j, policy->cpus) {
-		unsigned int tmp_idle_ticks;
+		unsigned int tmp_idle_ticks, total_idle_ticks;
 		struct cpu_dbs_info_s *j_dbs_info;
 
-		if (j == cpu)
-			continue;
-
 		j_dbs_info = &per_cpu(cpu_dbs_info, j);
 		/* Check for frequency increase */
 		total_idle_ticks = kstat_cpu(j).cpustat.idle +
@@ -330,7 +312,7 @@ static void dbs_check_cpu(int cpu)
 	down_idle_ticks = (100 - dbs_tuners_ins.down_threshold) *
 			sampling_rate_in_HZ(freq_down_sampling_rate);
 
-	if (idle_ticks > down_idle_ticks ) {
+	if (idle_ticks > down_idle_ticks) {
 		freq_down_step = (5 * policy->max) / 100;
 
 		/* max freq cannot be less than 100. But who knows.... */

[-- Attachment #4: ondemand-save-idle-up-for-all-cpu-2.6.11.patch --]
[-- Type: text/x-patch, Size: 587 bytes --]

--- linux-2.6.11/drivers/cpufreq/cpufreq_ondemand.c.clean2	2005-03-06 19:04:12.000000000 +0100
+++ linux-2.6.11/drivers/cpufreq/cpufreq_ondemand.c	2005-03-06 19:09:52.000000000 +0100
@@ -277,7 +277,12 @@
 		__cpufreq_driver_target(policy, policy->max, 
 			CPUFREQ_RELATION_H);
 		down_skip[cpu] = 0;
-		this_dbs_info->prev_cpu_idle_down = this_dbs_info->prev_cpu_idle_up;
+		for_each_cpu_mask(j, policy->cpus) {
+			struct cpu_dbs_info_s *j_dbs_info;
+
+			j_dbs_info = &per_cpu(cpu_dbs_info, j);
+			j_dbs_info->prev_cpu_idle_down = j_dbs_info->prev_cpu_idle_up;
+		}
 		return;
 	}
 

Re: cpufreq on-demand governor up_treshold?

[Bruno Ducrot]

On Mon, Mar 14, 2005 at 08:57:52AM +0100, Eric Piel wrote:
> BTW, DaveJ, Dominik, I couldn't find them in the daily-snapshot 
> available at codemonkey.org.uk. Should I worry, or is it just due to 
> some latency between your private trees and the public one?
> 

This happens those days only when I upgrade the LINUX_2_4 branch
(and only because its easier for me to diff between HEAD and LINUX_2_4).

-- 
Bruno Ducrot

--  Which is worse:  ignorance or apathy?
--  Don't know.  Don't care.

Re: cpufreq on-demand governor up_treshold?

[Jan De Luyck]

On Monday 14 March 2005 08:57, Eric Piel wrote:
> Jan De Luyck a écrit :
> > Hello lists,
> >
> > (please cc me from cpufreq list)
> >
> > I've since yesterday started using the ondemand governor. Seems to work
> > fine, tho I can't seem to find a reason why it keeps scaling my processor
> > speed upwards tho the processor use never exceeds 30% (been watching top
> > -d 1).
> >
> >
> > Any hints?
>
> You can try the three attached patches in the order :
> ondemand-cleanup-factorise-idle-measurement-2.6.11.patch
> ondemand-save-idle-up-for-all-cpu-2.6.11.patch
> ondemand-automatic-downscaling-2.6.11-accepted.patch
>
> They are available on the cpufreq list but as it's difficult to access
> it I'm sending them again, all together. These are the last things that
> Venki and I have been working on. It should solve your problem
> (actually, only the last patch, but it depends on the two previous
> patches). Please, let me know if it works.

Okay, now the behaviour of the ondemand governor looks more 'sane'. Thanks, it 
looks like a huge improvement :)

Jan

-- 
Snow and adolescence are the only problems that disappear if you ignore
them long enough.

Re: cpufreq on-demand governor up_treshold?

[Dominik Brodowski]

On Mon, Mar 14, 2005 at 08:57:52AM +0100, Eric Piel wrote:
> BTW, DaveJ, Dominik, I couldn't find them in the daily-snapshot 
> available at codemonkey.org.uk. Should I worry, or is it just due to 
> some latency between your private trees and the public one?

/me has no official position wrt cpufreq core [except userspace
cpufrequtils, but I intend to hand this over to somebody else in the next
few months].

Dave, as maintainer of cpufreq, has a cpufreq bitkeeper tree [http interface
at http://linux-dj.bkbits.net/ ] which is exported as plain diff daily at
http://www.codemonkey.org.uk/projects/cpufreq/daily-snapshots/ . This does
not contain your patches yet, probably because he's still busy with other
stuff.

Thanks,
	Dominik

Re: cpufreq on-demand governor up_treshold?

[Dave Jones]

On Mon, Mar 14, 2005 at 08:57:52AM +0100, Eric Piel wrote:
 > Jan De Luyck a écrit :
 > >Hello lists,
 > >
 > >(please cc me from cpufreq list)
 > >
 > >I've since yesterday started using the ondemand governor. Seems to work 
 > >fine, tho I can't seem to find a reason why it keeps scaling my processor 
 > >speed upwards tho the processor use never exceeds 30% (been watching top 
 > >-d 1). 
 > :
 > :
 > >Any hints?
 > You can try the three attached patches in the order :
 > ondemand-cleanup-factorise-idle-measurement-2.6.11.patch
 > ondemand-save-idle-up-for-all-cpu-2.6.11.patch
 > ondemand-automatic-downscaling-2.6.11-accepted.patch
 > 
 > They are available on the cpufreq list but as it's difficult to access 
 > it I'm sending them again, all together. These are the last things that 
 > Venki and I have been working on. It should solve your problem 
 > (actually, only the last patch, but it depends on the two previous 
 > patches). Please, let me know if it works.
 > 
 > BTW, DaveJ, Dominik, I couldn't find them in the daily-snapshot 
 > available at codemonkey.org.uk. Should I worry, or is it just due to 
 > some latency between your private trees and the public one?

I'm preparing the first cpufreq->linus sync right now.
Can you write up some descriptions & signed-off-by: lines for
these three please ?

Thanks,
		Dave