Skip to contentSkip to conversation

Does PowerTop support MHz numbers for Intel Xeon L5410 CPUs ?

Powertop Valentin Hoebel 3 participants 12 messages Jul 3, 2014 – Jul 9, 2014

lore
Share & tools
lore

Valentin Hoebel

author
#0 lore raw

Hello,

I have been using PowerTop for some time now and would like to thank all contributors for their work, especially Intel!

However, when using PowerTop on Debian 7 (Kernel 3.2, PowerTop package 2.0-0.3) and navigating to "Frequency stats", it does not show the MHz numbers. Instead, it shows me the following:

CPU 0
Idle 138%

CPU 2
Idle 111,7%

CPU 1
Idle 111,7%

CPU 3
Idle 111,7%

Is the CPU model Intel Xeon L5410 simply not supported or is there no way to read out the frequencies?

Browsing the source code, I did not stumple accross this CPU model.

Thanks in advance.

Best regards
Valentin

Signature · 18 lines
Mit freundlichen Grüßen

Valentin Höbel
Cloud Architect

NFON AG
Leonrodstraße 68
D-80636 München

Telefon +49 89 45300-407
valentin.hoebel(a)nfon.com
www.nfon.com

Support-Hotline der NFON AG
Telefon 0800 - 6366 555

Die gesetzlichen Pflichtangaben finden Sie unter 
https://www.nfon.net/pflichtangaben
#1 lore raw
Quoted 8 lines · Valentin Hoebel

Hello,

I have been using PowerTop for some time now and would like to thank all contributors for their work, especially Intel!

However, when using PowerTop on Debian 7 (Kernel 3.2, PowerTop package 2.0-0.3) and navigating to "Frequency stats", it does not show the MHz numbers. Instead, it shows me the following:

Hello,
am not really familiar with debian versioning, but if "2.0" means upstream release v2.0, then you use a really-really outdated powertop (several years). in that case I'd recommend to switch to recent version (2.6 IIRC).

	-ss
Quoted 47 lines

CPU 0
Idle 138%

CPU 2
Idle 111,7%

CPU 1
Idle 111,7%

CPU 3
Idle 111,7%

Is the CPU model Intel Xeon L5410 simply not supported or is there no way to read out the frequencies?

Browsing the source code, I did not stumple accross this CPU model.

Thanks in advance.

Best regards
Valentin

Signature · 22 lines
Mit freundlichen Grüßen

Valentin Höbel
Cloud Architect

NFON AG
Leonrodstraße 68
D-80636 München

Telefon +49 89 45300-407
valentin.hoebel(a)nfon.com
www.nfon.com

Support-Hotline der NFON AG
Telefon 0800 - 6366 555

Die gesetzlichen Pflichtangaben finden Sie unter
https://www.nfon.net/pflichtangaben
_______________________________________________
PowerTop mailing list
PowerTop(a)lists.01.org
https://lists.01.org/mailman/listinfo/powertop
author
#2 lore raw

Hi,

as you already guessed Debian still uses a 2.0 version. However, I git cloned 2.6.1 and the "head" versions from GitHub, built them both and tried them - still the same result.

Is it known if this particular CPU model does not support reading out the CPU frequency values or could this be another problem?

Best regards
Valentin

Am 03.07.2014 15:59, schrieb Sergey Senozhatsky:

Quoted 65 lines · Valentin Hoebel
Quoted 8 lines

Hello,

I have been using PowerTop for some time now and would like to thank all contributors for their work, especially Intel!

However, when using PowerTop on Debian 7 (Kernel 3.2, PowerTop package 2.0-0.3) and navigating to "Frequency stats", it does not show the MHz numbers. Instead, it shows me the following:

Hello,
am not really familiar with debian versioning, but if "2.0" means upstream release v2.0, then you use a really-really outdated powertop (several years). in that case I'd recommend to switch to recent version (2.6 IIRC).

	-ss
Quoted 47 lines

CPU 0
Idle 138%

CPU 2
Idle 111,7%

CPU 1
Idle 111,7%

CPU 3
Idle 111,7%

Is the CPU model Intel Xeon L5410 simply not supported or is there no way to read out the frequencies?

Browsing the source code, I did not stumple accross this CPU model.

Thanks in advance.

Best regards
Valentin

--
Mit freundlichen Grüßen

Valentin Höbel
Cloud Architect

NFON AG
Leonrodstraße 68
D-80636 München

Telefon +49 89 45300-407
valentin.hoebel(a)nfon.com
www.nfon.com

Support-Hotline der NFON AG
Telefon 0800 - 6366 555

Die gesetzlichen Pflichtangaben finden Sie unter
https://www.nfon.net/pflichtangaben
_______________________________________________
PowerTop mailing list
PowerTop(a)lists.01.org
https://lists.01.org/mailman/listinfo/powertop

Signature · 18 lines
Mit freundlichen Grüßen

Valentin Höbel
Cloud Architect

NFON AG
Leonrodstraße 68
D-80636 München

Telefon +49 89 45300-407
valentin.hoebel(a)nfon.com
www.nfon.com

Support-Hotline der NFON AG
Telefon 0800 - 6366 555

Die gesetzlichen Pflichtangaben finden Sie unter 
https://www.nfon.net/pflichtangaben
#3 lore raw
Quoted 8 lines · Valentin Hoebel

Hi,

as you already guessed Debian still uses a 2.0 version. However, I git cloned 2.6.1 and the "head" versions from GitHub, built them both and tried them - still the same result.

Is it known if this particular CPU model does not support reading out the CPU frequency values or could this be another problem?

hm... a *wild* guess. can you check what cpu model is getting passed to nhm_core::nhm_core(int model)? the problem might be (or might be not) that we don't set any of C-state flags. if none of C-states set, then no MSR work is getting done, and hence nothing is reported back. (yeah... we could check documentation, search the web, etc. but let's do it the lazy way).

something like this

NOTE:
NOT TESTED, NOT EVEN COMPILE-TESTED. NOT A PATCH, NOT A FIX. FOR TESTING PURPOSE ONLY.

---

 src/cpu/intel_cpus.cpp | 7 +++++++
 1 file changed, 7 insertions(+)
src/cpu/intel_cpus.cpp +7 -0
index e3cdaf2..ae083cf 100644
@@ -64,10 +64,13 @@ int is_supported_intel_cpu(int model)
 {
 	int i;
 
+	fprintf(stderr, "check cpu model 0x%X\n", model);
+
 	for (i = 0; intel_cpu_models[i] != 0; i++)
 		if (model == intel_cpu_models[i])
 			return 1;
 
+	printf(stderr, "ERROR: not supported\n");
 	return 0;
 }
 
@@ -91,6 +94,8 @@ nhm_core::nhm_core(int model)
 {
 	has_c2c7_res = 0;
 
+	fprintf(stderr, "nhm_core model: 0x%X\n", model);
+
 	switch(model) {
 		case 0x2A:	/* SNB */
 		case 0x2D:	/* SNB Xeon */
@@ -260,6 +265,8 @@ nhm_package::nhm_package(int model)
 	has_c8c9c10_res = 0;
 	has_c2c7_res = 0;
 
+	fprintf(stderr, "nhm_package model: 0x%X\n", model);
+
 	switch(model) {
 		case 0x2A:	/* SNB */
 		case 0x2D:	/* SNB Xeon */
#4 lore raw
Quoted 10 lines · Sergey Senozhatsky
Quoted 8 lines · Valentin Hoebel

Hi,

as you already guessed Debian still uses a 2.0 version. However, I git cloned 2.6.1 and the "head" versions from GitHub, built them both and tried them - still the same result.

Is it known if this particular CPU model does not support reading out the CPU frequency values or could this be another problem?

Quoted 9 lines

hm... a *wild* guess. can you check what cpu model is getting passed to nhm_core::nhm_core(int model)? the problem might be (or might be not) that we don't set any of C-state flags. if none of C-states set, then no MSR work is getting done, and hence nothing is reported back. (yeah... we could check documentation, search the web, etc. but let's do it the lazy way).

something like this

oh. the patch is against the current upstream (iow, git) powertop source code.

	-ss
Quoted 46 lines

NOTE:
NOT TESTED, NOT EVEN COMPILE-TESTED. NOT A PATCH, NOT A FIX. FOR TESTING PURPOSE ONLY.

---

 src/cpu/intel_cpus.cpp | 7 +++++++
 1 file changed, 7 insertions(+)
src/cpu/intel_cpus.cpp +7 -0
index e3cdaf2..ae083cf 100644
@@ -64,10 +64,13 @@ int is_supported_intel_cpu(int model)
 {
 	int i;
 
+	fprintf(stderr, "check cpu model 0x%X\n", model);
+
 	for (i = 0; intel_cpu_models[i] != 0; i++)
 		if (model == intel_cpu_models[i])
 			return 1;
 
+	printf(stderr, "ERROR: not supported\n");
 	return 0;
 }
 
@@ -91,6 +94,8 @@ nhm_core::nhm_core(int model)
 {
 	has_c2c7_res = 0;
 
+	fprintf(stderr, "nhm_core model: 0x%X\n", model);
+
 	switch(model) {
 		case 0x2A:	/* SNB */
 		case 0x2D:	/* SNB Xeon */
@@ -260,6 +265,8 @@ nhm_package::nhm_package(int model)
 	has_c8c9c10_res = 0;
 	has_c2c7_res = 0;
 
+	fprintf(stderr, "nhm_package model: 0x%X\n", model);
+
 	switch(model) {
 		case 0x2A:	/* SNB */
 		case 0x2D:	/* SNB Xeon */
author
#5 lore raw

Hi,

thanks a lot for your efforts, I appreciate it.
My co-worker applied your patch and sent me the following output:

--------------------------------------------------
/ ./powertop -t 1 --csv
⏎
Loaded 0 prior measurements
Cannot load from file /var/cache/powertop/saved_parameters.powertop check cpu model 0x17
ERROR: not supported
RAPL device for cpu 0
RAPL device for cpu 0
check cpu model 0x17
ERROR: not supported
check cpu model 0x17
ERROR: not supported
check cpu model 0x17
ERROR: not supported
check cpu model 0x17
ERROR: not supported
check cpu model 0x17
ERROR: not supported
check cpu model 0x17
ERROR: not supported
check cpu model 0x17
ERROR: not supported
check cpu model 0x17
ERROR: not supported
Cannot load from file /var/cache/powertop/saved_parameters.powertop Preparing to take measurements
unknown op '{'
Taking 1 measurement(s) for a duration of 1 second(s) each. PowerTOP outputing using base filename powertop.csv

\ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

I hope this helps.

Best regards
Valentin

Am 07.07.2014 12:28, schrieb Sergey Senozhatsky:

Quoted 71 lines · Sergey Senozhatsky
Quoted 10 lines · Valentin Hoebel
Quoted 8 lines

Hi,

as you already guessed Debian still uses a 2.0 version. However, I git cloned 2.6.1 and the "head" versions from GitHub, built them both and tried them - still the same result.

Is it known if this particular CPU model does not support reading out the CPU frequency values or could this be another problem?

Quoted 9 lines

hm... a *wild* guess. can you check what cpu model is getting passed to nhm_core::nhm_core(int model)? the problem might be (or might be not) that we don't set any of C-state flags. if none of C-states set, then no MSR work is getting done, and hence nothing is reported back. (yeah... we could check documentation, search the web, etc. but let's do it the lazy way).

something like this

oh. the patch is against the current upstream (iow, git) powertop source code.

	-ss
Quoted 46 lines

NOTE:
NOT TESTED, NOT EVEN COMPILE-TESTED. NOT A PATCH, NOT A FIX. FOR TESTING PURPOSE ONLY.

---

  src/cpu/intel_cpus.cpp | 7 +++++++
  1 file changed, 7 insertions(+)
src/cpu/intel_cpus.cpp +7 -0
index e3cdaf2..ae083cf 100644
@@ -64,10 +64,13 @@ int is_supported_intel_cpu(int model)
  {
  	int i;
 
+	fprintf(stderr, "check cpu model 0x%X\n", model);
+
  	for (i = 0; intel_cpu_models[i] != 0; i++)
  		if (model == intel_cpu_models[i])
  			return 1;
 
+	printf(stderr, "ERROR: not supported\n");
  	return 0;
  }
 
@@ -91,6 +94,8 @@ nhm_core::nhm_core(int model)
  {
  	has_c2c7_res = 0;
 
+	fprintf(stderr, "nhm_core model: 0x%X\n", model);
+
  	switch(model) {
  		case 0x2A:	/* SNB */
  		case 0x2D:	/* SNB Xeon */
@@ -260,6 +265,8 @@ nhm_package::nhm_package(int model)
  	has_c8c9c10_res = 0;
  	has_c2c7_res = 0;
 
+	fprintf(stderr, "nhm_package model: 0x%X\n", model);
+
  	switch(model) {
  		case 0x2A:	/* SNB */
  		case 0x2D:	/* SNB Xeon */
Signature · 18 lines
Mit freundlichen Grüßen

Valentin Höbel
Cloud Architect

NFON AG
Leonrodstraße 68
D-80636 München

Telefon +49 89 45300-407
valentin.hoebel(a)nfon.com
www.nfon.com

Support-Hotline der NFON AG
Telefon 0800 - 6366 555

Die gesetzlichen Pflichtangaben finden Sie unter 
https://www.nfon.net/pflichtangaben
#6 lore raw
Quoted 9 lines · Valentin Hoebel

Hi,

thanks a lot for your efforts, I appreciate it.
My co-worker applied your patch and sent me the following output:

--------------------------------------------------
/ ./powertop -t 1 --csv
⏎

thanks! ok, so cpu model 0x17 is not listed as a supported one in src/cpu/intel_cpus.cpp:intel_cpu_models[] to begin with, and no C-states are initialized as a result. Let's wait for Intel devs to add cpu/core/package models and corresponding C-states.

Alexandra, will you take care of this?

	-ss
Quoted 130 lines

Loaded 0 prior measurements
Cannot load from file /var/cache/powertop/saved_parameters.powertop check cpu model 0x17
ERROR: not supported
RAPL device for cpu 0
RAPL device for cpu 0
check cpu model 0x17
ERROR: not supported
check cpu model 0x17
ERROR: not supported
check cpu model 0x17
ERROR: not supported
check cpu model 0x17
ERROR: not supported
check cpu model 0x17
ERROR: not supported
check cpu model 0x17
ERROR: not supported
check cpu model 0x17
ERROR: not supported
check cpu model 0x17
ERROR: not supported
Cannot load from file /var/cache/powertop/saved_parameters.powertop Preparing to take measurements
unknown op '{'
Taking 1 measurement(s) for a duration of 1 second(s) each. PowerTOP outputing using base filename powertop.csv

\ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

I hope this helps.

Best regards
Valentin

Am 07.07.2014 12:28, schrieb Sergey Senozhatsky:

Quoted 71 lines · Sergey Senozhatsky
Quoted 10 lines · Valentin Hoebel
Quoted 8 lines

Hi,

as you already guessed Debian still uses a 2.0 version. However, I git cloned 2.6.1 and the "head" versions from GitHub, built them both and tried them - still the same result.

Is it known if this particular CPU model does not support reading out the CPU frequency values or could this be another problem?

Quoted 9 lines

hm... a *wild* guess. can you check what cpu model is getting passed to nhm_core::nhm_core(int model)? the problem might be (or might be not) that we don't set any of C-state flags. if none of C-states set, then no MSR work is getting done, and hence nothing is reported back. (yeah... we could check documentation, search the web, etc. but let's do it the lazy way).

something like this

oh. the patch is against the current upstream (iow, git) powertop source code.

	-ss
Quoted 46 lines

NOTE:
NOT TESTED, NOT EVEN COMPILE-TESTED. NOT A PATCH, NOT A FIX. FOR TESTING PURPOSE ONLY.

---

 src/cpu/intel_cpus.cpp | 7 +++++++
 1 file changed, 7 insertions(+)
src/cpu/intel_cpus.cpp +7 -0
index e3cdaf2..ae083cf 100644
@@ -64,10 +64,13 @@ int is_supported_intel_cpu(int model)
 {
 	int i;
 
+	fprintf(stderr, "check cpu model 0x%X\n", model);
+
 	for (i = 0; intel_cpu_models[i] != 0; i++)
 		if (model == intel_cpu_models[i])
 			return 1;
 
+	printf(stderr, "ERROR: not supported\n");
 	return 0;
 }
 
@@ -91,6 +94,8 @@ nhm_core::nhm_core(int model)
 {
 	has_c2c7_res = 0;
 
+	fprintf(stderr, "nhm_core model: 0x%X\n", model);
+
 	switch(model) {
 		case 0x2A:	/* SNB */
 		case 0x2D:	/* SNB Xeon */
@@ -260,6 +265,8 @@ nhm_package::nhm_package(int model)
 	has_c8c9c10_res = 0;
 	has_c2c7_res = 0;
 
+	fprintf(stderr, "nhm_package model: 0x%X\n", model);
+
 	switch(model) {
 		case 0x2A:	/* SNB */
 		case 0x2D:	/* SNB Xeon */
Signature · 18 lines
Mit freundlichen Grüßen

Valentin Höbel
Cloud Architect

NFON AG
Leonrodstraße 68
D-80636 München

Telefon +49 89 45300-407
valentin.hoebel(a)nfon.com
www.nfon.com

Support-Hotline der NFON AG
Telefon 0800 - 6366 555

Die gesetzlichen Pflichtangaben finden Sie unter
https://www.nfon.net/pflichtangaben
#7 lore raw
Quoted 11 lines · Sergey Senozhatsky
Quoted 9 lines · Valentin Hoebel

Hi,

thanks a lot for your efforts, I appreciate it.
My co-worker applied your patch and sent me the following output:

   --------------------------------------------------

/ ./powertop -t 1 --csv
⏎

Quoted 16 lines

thanks! ok, so cpu model 0x17 is not listed as a supported one in src/cpu/intel_cpus.cpp:intel_cpu_models[] to begin with, and no C-states are initialized as a result. Let's wait for Intel devs to add cpu/core/package models and corresponding C-states.

Alexandra, will you take care of this?

	-ss
Quoted 7 lines

Loaded 0 prior measurements
Cannot load from file /var/cache/powertop/saved_parameters.powertop check cpu model 0x17
ERROR: not supported
RAPL device for cpu 0
RAPL device for cpu 0
check cpu model 0x17

so turns out the L5410 is one generation prior to NHM, and does not support some of the internal MSRs we're using. I'll talk to Alexandra (pending vacations, it's that season) to see what we can do

#8 lore raw
Quoted 27 lines · Arjan van de Ven
Quoted 10 lines
Quoted 9 lines

Hi,

thanks a lot for your efforts, I appreciate it.
My co-worker applied your patch and sent me the following output:

--------------------------------------------------
/ ./powertop -t 1 --csv
⏎

Quoted 16 lines

thanks! ok, so cpu model 0x17 is not listed as a supported one in src/cpu/intel_cpus.cpp:intel_cpu_models[] to begin with, and no C-states are initialized as a result. Let's wait for Intel devs to add cpu/core/package models and corresponding C-states.

Alexandra, will you take care of this?

	-ss
Quoted 7 lines

Loaded 0 prior measurements
Cannot load from file /var/cache/powertop/saved_parameters.powertop check cpu model 0x17
ERROR: not supported
RAPL device for cpu 0
RAPL device for cpu 0
check cpu model 0x17

Quoted 3 lines

so turns out the L5410 is one generation prior to NHM, and does not support some of the internal MSRs we're using. I'll talk to Alexandra (pending vacations, it's that season) to see what we can do

not sure that powertop performs any specific MSRs at all, since most of MSRs are called correspondingly to c-state flags (has_cXXX), e.g.

__init:

	if (model == 0x45 || model ==0x3D)
		has_c8c9c10_res = 1;

and later:

	if (has_c8c9c10_res) {
		c8_before    = get_msr(number, MSR_PKG_C8_RESIDENCY);
		c9_before    = get_msr(number, MSR_PKG_C9_RESIDENCY);
		c10_before    = get_msr(number, MSR_PKG_C10_RESIDENCY);
	}

	and so on.

	-ss
#9 lore raw
Quoted 12 lines

not sure that powertop performs any specific MSRs at all, since most of MSRs are called correspondingly to c-state flags (has_cXXX), e.g.

__init:

	if (model == 0x45 || model ==0x3D)
		has_c8c9c10_res = 1;

and later:

Quoted 7 lines
	if (has_c8c9c10_res) {
		c8_before    = get_msr(number, MSR_PKG_C8_RESIDENCY);
		c9_before    = get_msr(number, MSR_PKG_C9_RESIDENCY);
		c10_before    = get_msr(number, MSR_PKG_C10_RESIDENCY);
	}

	and so on.

but if the set is empty, you don't get any data, and the "emulation" side of powertop is likely a better choice... since there are C states, just not counters.

#10 lore raw
Quoted 17 lines · Arjan van de Ven
Quoted 17 lines

not sure that powertop performs any specific MSRs at all, since most of MSRs are called correspondingly to c-state flags (has_cXXX), e.g.

__init:

	if (model == 0x45 || model ==0x3D)
		has_c8c9c10_res = 1;

and later:

	if (has_c8c9c10_res) {
		c8_before    = get_msr(number, MSR_PKG_C8_RESIDENCY);
		c9_before    = get_msr(number, MSR_PKG_C9_RESIDENCY);
		c10_before    = get_msr(number, MSR_PKG_C10_RESIDENCY);
	}
Quoted 6 lines
Quoted 1 line

and so on.

but if the set is empty, you don't get any data, and the "emulation" side of powertop is likely a better choice... since there are C states, just not counters.

looks like powertop checks C6 state get_msr(number, MSR_PKG_C6_RESIDENCY)' in start and end measurement for any intel cpu, along with get_msr(first_cpu, MSR_TSC)'. anyway, need to investigate. the output that Valentin sees is rather strange, saying that cpu0 is 138% Idle.

	          CPU 0
	Idle      138%

	          CPU 2
	Idle      111,7%

	          CPU 1
	Idle      111,7%

	          CPU 3
	Idle      111,7%

	-ss
author
#11 lore raw

Hi,

not sure if this is an important information or not, but I am getting the "138% idle" behaviour for all of the systems where reading out the current CPU frequencies via PowerTop is not possible.

Best regards
Valentin

Am 08.07.2014 13:05, schrieb Sergey Senozhatsky:

Quoted 44 lines · Arjan van de Ven
Quoted 17 lines
Quoted 17 lines

not sure that powertop performs any specific MSRs at all, since most of MSRs are called correspondingly to c-state flags (has_cXXX), e.g.

__init:

	if (model == 0x45 || model ==0x3D)
		has_c8c9c10_res = 1;

and later:

	if (has_c8c9c10_res) {
		c8_before    = get_msr(number, MSR_PKG_C8_RESIDENCY);
		c9_before    = get_msr(number, MSR_PKG_C9_RESIDENCY);
		c10_before    = get_msr(number, MSR_PKG_C10_RESIDENCY);
	}
Quoted 6 lines
Quoted 1 line

and so on.

but if the set is empty, you don't get any data, and the "emulation" side of powertop is likely a better choice... since there are C states, just not counters.

looks like powertop checks C6 state get_msr(number, MSR_PKG_C6_RESIDENCY)' in start and end measurement for any intel cpu, along with get_msr(first_cpu, MSR_TSC)'. anyway, need to investigate. the output that Valentin sees is rather strange, saying that cpu0 is 138% Idle.

	          CPU 0
	Idle      138%

	          CPU 2
	Idle      111,7%

	          CPU 1
	Idle      111,7%

	          CPU 3
	Idle      111,7%

	-ss
Signature · 18 lines
Mit freundlichen Grüßen

Valentin Höbel
Cloud Architect

NFON AG
Leonrodstraße 68
D-80636 München

Telefon +49 89 45300-407
valentin.hoebel(a)nfon.com
www.nfon.com

Support-Hotline der NFON AG
Telefon 0800 - 6366 555

Die gesetzlichen Pflichtangaben finden Sie unter 
https://www.nfon.net/pflichtangaben