Intel Clovertown vs. frequency scaling

Intel Clovertown vs. frequency scaling

[Andras.Horvath]

Hi,

I have a number of Clovertown (intel quadcore) machines and, even with
the latest 2.6.23, it seems to me that frequency scaling is not
supported (e.g. the /sys/devices/system/cpu/cpu0/cpufreq/* stuff does
not appear even though CONFIG_X86_SPEEDSTEP_CENTRINO is "y").

Is that as expected?

This is what the CPU (one core in one of the two sockets) looks like:

# head -24 /proc/cpuinfo 
processor       : 0
vendor_id       : GenuineIntel
cpu family      : 6
model           : 15
model name      : Intel(R) Xeon(R) CPU           L5335  @ 2.00GHz
stepping        : 11
cpu MHz         : 1995.105
cache size      : 4096 KB
physical id     : 0
siblings        : 4
core id         : 0
cpu cores       : 4
fdiv_bug        : no
hlt_bug         : no
f00f_bug        : no
coma_bug        : no
fpu             : yes
fpu_exception   : yes
cpuid level     : 10
wp              : yes
flags           : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe nx lm constant_tsc arch_perfmon pebs bts pni monitor ds_cpl vmx tm2 ssse3 cx16 xtpr dca lahf_lm
bogomips        : 3994.09
clflush size    : 64


also:

# cat /proc/acpi/processor/CPU0/info 
processor id:            0
acpi id:                 0
bus mastering control:   no
power management:        yes
throttling control:      yes
limit interface:         yes

# cat /proc/acpi/processor/CPU0/power 
active state:            C2
max_cstate:              C8
bus master activity:     00000000
maximum allowed latency: 20000 usec
states:
    C1:                  type[C1] promotion[C2] demotion[--] latency[020] usage[00003940] duration[00000000000000000000]
   *C2:                  type[C2] promotion[--] demotion[C1] latency[040] usage[00013668] duration[00000000000359543325]


FWIW, I also have some other Core-based (Woodcrest, Intel L5160) systems
and cpufreq works fine there.

Any help or insight would be appreciated. 

(I'm experimenting with ways to lower the average power consumption of a
large number of servers, working my way through http://www.lesswatts.org
etc.  I'm willing to patch kernels, run tests, and I have proper power
meters to check the results.)

Andras

Re: Intel Clovertown vs. frequency scaling

[Jarod Wilson]

[-- Attachment #1.1: Type: text/plain, Size: 521 bytes --]

Andras.Horvath@cern.ch wrote:
> I have a number of Clovertown (intel quadcore) machines and, even with
> the latest 2.6.23, it seems to me that frequency scaling is not
> supported (e.g. the /sys/devices/system/cpu/cpu0/cpufreq/* stuff does
> not appear even though CONFIG_X86_SPEEDSTEP_CENTRINO is "y").

I believe you want acpi-cpufreq, not speedstep-centrino. Works just fine
for my quad-core systems, anyhow. If that doesn't work, its probably the
bios at fault.

-- 
Jarod Wilson
jwilson@redhat.com



[-- Attachment #1.2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 251 bytes --]

[-- Attachment #2: Type: text/plain, Size: 147 bytes --]

_______________________________________________
Cpufreq mailing list
Cpufreq@lists.linux.org.uk
http://lists.linux.org.uk/mailman/listinfo/cpufreq

Re: Intel Clovertown vs. frequency scaling

[Stephen J. Gowdy]

[-- Attachment #1: Type: TEXT/PLAIN, Size: 741 bytes --]

I have a quad-core version that uses the centrino one (E5345). However, it 
only has two available speeds, 2GHz and 2.33GHz. Do you get more options?

[pcphuat27] ~ > cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_driver
centrino
[pcphuat27] ~ > cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies
2333000 2000000


--
  /------------------------------------+-------------------------\
|Stephen J. Gowdy, SLAC               | CERN     Office: 32-2-A22|
|http://www.slac.stanford.edu/~gowdy/ | CH-1211 Geneva 23        |
|                                     | Switzerland              |
|EMail: gowdy@slac.stanford.edu       | Tel: +41 22 767 5840     |
  \------------------------------------+-------------------------/

[-- Attachment #2.1: Type: TEXT/PLAIN, Size: 551 bytes --]

Andras.Horvath@cern.ch wrote:

> I have a number of Clovertown (intel quadcore) machines and, even with

> the latest 2.6.23, it seems to me that frequency scaling is not

> supported (e.g. the /sys/devices/system/cpu/cpu0/cpufreq/* stuff does

> not appear even though CONFIG_X86_SPEEDSTEP_CENTRINO is "y").



I believe you want acpi-cpufreq, not speedstep-centrino. Works just fine

for my quad-core systems, anyhow. If that doesn't work, its probably the

bios at fault.



-- 

Jarod Wilson

jwilson@redhat.com






[-- Attachment #2.2: OpenPGP digital signature --]
[-- Type: APPLICATION/PGP-SIGNATURE, Size: 251 bytes --]

[-- Attachment #3: Type: TEXT/PLAIN, Size: 151 bytes --]

_______________________________________________

Cpufreq mailing list

Cpufreq@lists.linux.org.uk

http://lists.linux.org.uk/mailman/listinfo/cpufreq


[-- Attachment #4: Type: text/plain, Size: 147 bytes --]

_______________________________________________
Cpufreq mailing list
Cpufreq@lists.linux.org.uk
http://lists.linux.org.uk/mailman/listinfo/cpufreq

Re: Intel Clovertown vs. frequency scaling

[Jarod Wilson]

[-- Attachment #1.1: Type: text/plain, Size: 1301 bytes --]

Stephen J. Gowdy wrote:
> I have a quad-core version that uses the centrino one (E5345). However,
> it only has two available speeds, 2GHz and 2.33GHz. Do you get more
> options?
> 
> [pcphuat27] ~ > cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_driver
> centrino
> [pcphuat27] ~ > cat
> /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies
> 2333000 2000000

Not a heck of a lot more, but a bit more.

[jwilson@xeus ~]$ cat sys/devices/system/cpu/cpu0/cpufreq/scaling_driver
acpi-cpufreq
[jwilson@xeus ~]$ cat
/sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies
2660000 2128000 1596000

FYI, unless I'm mistaken, the speedstep-centrino driver is supposed to
be getting deprecated in favor of acpi-cpufreq.


> Andras.Horvath@cern.ch wrote:
>> I have a number of Clovertown (intel quadcore) machines and, even with
>> the latest 2.6.23, it seems to me that frequency scaling is not
>> supported (e.g. the /sys/devices/system/cpu/cpu0/cpufreq/* stuff does
>> not appear even though CONFIG_X86_SPEEDSTEP_CENTRINO is "y").
> 
> I believe you want acpi-cpufreq, not speedstep-centrino. Works just fine
> for my quad-core systems, anyhow. If that doesn't work, its probably the
> bios at fault.


-- 
Jarod Wilson
jwilson@redhat.com



[-- Attachment #1.2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 251 bytes --]

[-- Attachment #2: Type: text/plain, Size: 147 bytes --]

_______________________________________________
Cpufreq mailing list
Cpufreq@lists.linux.org.uk
http://lists.linux.org.uk/mailman/listinfo/cpufreq

RE: Intel Clovertown vs. frequency scaling

[Pallipadi, Venkatesh]

With latest kernels (2.6.22 and beyond) you should use acpi-cpufreq and
not speedstep-centrino.

The frequencies supported depends on bus speed and voltages supported by
the CPU. I think the thing here is you have a low voltage part which is
already running at lowest possible voltage and does not support any more
frequency/voltage reduction.

You mentioned that cpufreq works fine on Woodcrest, Intel L5160.
What are the freqs supported and what is the max freq for this part?

Thanks,
Venki
 
>-----Original Message-----
>From: cpufreq-bounces@lists.linux.org.uk 
>[mailto:cpufreq-bounces@lists.linux.org.uk] On Behalf Of 
>Andras.Horvath@cern.ch
>Sent: Thursday, October 11, 2007 8:07 AM
>To: cpufreq@lists.linux.org.uk
>Subject: Intel Clovertown vs. frequency scaling
>
>Hi,
>
>I have a number of Clovertown (intel quadcore) machines and, even with
>the latest 2.6.23, it seems to me that frequency scaling is not
>supported (e.g. the /sys/devices/system/cpu/cpu0/cpufreq/* stuff does
>not appear even though CONFIG_X86_SPEEDSTEP_CENTRINO is "y").
>
>Is that as expected?
>
>This is what the CPU (one core in one of the two sockets) looks like:
>
># head -24 /proc/cpuinfo 
>processor       : 0
>vendor_id       : GenuineIntel
>cpu family      : 6
>model           : 15
>model name      : Intel(R) Xeon(R) CPU           L5335  @ 2.00GHz
>stepping        : 11
>cpu MHz         : 1995.105
>cache size      : 4096 KB
>physical id     : 0
>siblings        : 4
>core id         : 0
>cpu cores       : 4
>fdiv_bug        : no
>hlt_bug         : no
>f00f_bug        : no
>coma_bug        : no
>fpu             : yes
>fpu_exception   : yes
>cpuid level     : 10
>wp              : yes
>flags           : fpu vme de pse tsc msr pae mce cx8 apic sep 
>mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 
>ss ht tm pbe nx lm constant_tsc arch_perfmon pebs bts pni 
>monitor ds_cpl vmx tm2 ssse3 cx16 xtpr dca lahf_lm
>bogomips        : 3994.09
>clflush size    : 64
>
>
>also:
>
># cat /proc/acpi/processor/CPU0/info 
>processor id:            0
>acpi id:                 0
>bus mastering control:   no
>power management:        yes
>throttling control:      yes
>limit interface:         yes
>
># cat /proc/acpi/processor/CPU0/power 
>active state:            C2
>max_cstate:              C8
>bus master activity:     00000000
>maximum allowed latency: 20000 usec
>states:
>    C1:                  type[C1] promotion[C2] demotion[--] 
>latency[020] usage[00003940] duration[00000000000000000000]
>   *C2:                  type[C2] promotion[--] demotion[C1] 
>latency[040] usage[00013668] duration[00000000000359543325]
>
>
>FWIW, I also have some other Core-based (Woodcrest, Intel 
>L5160) systems
>and cpufreq works fine there.
>
>Any help or insight would be appreciated. 
>
>(I'm experimenting with ways to lower the average power 
>consumption of a
>large number of servers, working my way through 
>http://www.lesswatts.org
>etc.  I'm willing to patch kernels, run tests, and I have proper power
>meters to check the results.)
>
>Andras
>
>_______________________________________________
>Cpufreq mailing list
>Cpufreq@lists.linux.org.uk
>http://lists.linux.org.uk/mailman/listinfo/cpufreq
>

Re: Intel Clovertown vs. frequency scaling

[Andras.Horvath]

> With latest kernels (2.6.22 and beyond) you should use acpi-cpufreq and
> not speedstep-centrino.

ah, thanks. :-) I found that that too was compiled in (and in fact
that's what's driving the Woodcrest box as well).

> The frequencies supported depends on bus speed and voltages supported by
> the CPU. I think the thing here is you have a low voltage part which is
> already running at lowest possible voltage and does not support any more
> frequency/voltage reduction.

Hm. So there are no P* states available (always running at top speed)
but the CPU just goes to some C* state when it's idle?

This looks right (even if I find it strange) as the specs
(http://download.intel.com/design/Xeon/datashts/31556903.pdf ) say that
"Note: Not all Quad-Core Intel® Xeon® Processor 5300 Series are capable
of supporting Enhanced Intel SpeedStep Technology. More details on which
processor frequencies will support this feature will be provided in
future releases of the Quad-Core Intel® Xeon® Processor 5300 Series NDA
Specification Update when available."

Nevertheless http://processorfinder.intel.com/details.aspx?sSpec=SLAEN
says the L5335 has Enhanced Intel Speedstep support.

> You mentioned that cpufreq works fine on Woodcrest, Intel L5160.
> What are the freqs supported and what is the max freq for this part?

As I said I was mistaken and it's using acpi-cpufreq. However, this:

model name      : Intel(R) Xeon(R) CPU            5160  @ 3.00GHz

supports the following frequencies (2.6.23):

# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies 
2997000 2664000 2331000 1998000 


Also, I have found myself another Clovertown:
model name      : Intel(R) Xeon(R) CPU           E5345  @ 2.33GHz

and this seems not to support no frequency scaling, no C-states, only
thermal throttling. Can this be true? From the specs mentioned above I'd
have expected at least some sort of halt state.
http://processorfinder.intel.com/details.aspx?sSpec=SL9YL says 'enhanced
intel speedstep' and C1E support, same as Mr. L5335 above.

I'm getting more and more confused about this :)

Andras

RE: Intel Clovertown vs. frequency scaling

[Pallipadi, Venkatesh]

 

>-----Original Message-----
>From: cpufreq-bounces@lists.linux.org.uk 
>[mailto:cpufreq-bounces@lists.linux.org.uk] On Behalf Of 
>Andras.Horvath@cern.ch
>Sent: Friday, October 12, 2007 8:01 AM
>To: cpufreq@lists.linux.org.uk
>Subject: Re: Intel Clovertown vs. frequency scaling
>
>> With latest kernels (2.6.22 and beyond) you should use 
>acpi-cpufreq and
>> not speedstep-centrino.
>
>ah, thanks. :-) I found that that too was compiled in (and in fact
>that's what's driving the Woodcrest box as well).
>
>> The frequencies supported depends on bus speed and voltages 
>supported by
>> the CPU. I think the thing here is you have a low voltage 
>part which is
>> already running at lowest possible voltage and does not 
>support any more
>> frequency/voltage reduction.
>
>Hm. So there are no P* states available (always running at top speed)
>but the CPU just goes to some C* state when it's idle?

Correct. On idle CPU will consume much less power than top speed, 
as it will go to C-state.

>
>This looks right (even if I find it strange) as the specs
>(http://download.intel.com/design/Xeon/datashts/31556903.pdf ) say that
>"Note: Not all Quad-Core Intel(r) Xeon(r) Processor 5300 Series are
capable
>of supporting Enhanced Intel SpeedStep Technology. More 
>details on which
>processor frequencies will support this feature will be provided in
>future releases of the Quad-Core Intel(r) Xeon(r) Processor 5300 Series
NDA
>Specification Update when available."
>
>Nevertheless http://processorfinder.intel.com/details.aspx?sSpec=SLAEN
>says the L5335 has Enhanced Intel Speedstep support.
>

Yes. I noticed the processorfinder thing yesterday and trying to get
people to add clarification onto that page.

>> You mentioned that cpufreq works fine on Woodcrest, Intel L5160.
>> What are the freqs supported and what is the max freq for this part?
>
>As I said I was mistaken and it's using acpi-cpufreq. However, this:
>
>model name      : Intel(R) Xeon(R) CPU            5160  @ 3.00GHz
>
>supports the following frequencies (2.6.23):
>
># cat 
>/sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies 
>2997000 2664000 2331000 1998000 
>

That seems correct.

>Also, I have found myself another Clovertown:
>model name      : Intel(R) Xeon(R) CPU           E5345  @ 2.33GHz
>
>and this seems not to support no frequency scaling, no C-states, only
>thermal throttling. Can this be true? From the specs mentioned 
>above I'd
>have expected at least some sort of halt state.
>http://processorfinder.intel.com/details.aspx?sSpec=SL9YL says 
>'enhanced
>intel speedstep' and C1E support, same as Mr. L5335 above.
>

Hmmm. This processor should support 2GHz frequency. Most probably BIOS
is not providing information for the driver to enable speedstep here.
Can you make sure you have the latest BIOS and also check whether there
are any options to enable speedstep in BIOS setup. Send me the output of
acpidump (you can find acpidump in latest pmtools package here -
http://www.kernel.org/pub/linux/kernel/people/lenb/acpi/utils/ ). I can
take a look at what BIOS is saying about speedstep.

Thanks,
Venki

Re: Intel Clovertown vs. frequency scaling

[Andras.Horvath]

(sorry for the late reply)

> Hmmm. This processor should support 2GHz frequency. Most probably BIOS
> is not providing information for the driver to enable speedstep here.

This is embarassing, the option (named "EIST support") was off. Turning
it on resulted in

# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies 
2333000 2000000 

but no C-state info (like on the L5335) with 2.6.23 and acpi-cpufreq.

> Can you make sure you have the latest BIOS and also check whether there
> are any options to enable speedstep in BIOS setup. Send me the output of
> acpidump (you can find acpidump in latest pmtools package here -
> http://www.kernel.org/pub/linux/kernel/people/lenb/acpi/utils/ ). I can
> take a look at what BIOS is saying about speedstep.

Hm... I may be missing something again, but acpidump doesn't find the
RSD PTR table, and dmesg doesn't help either (all ACPI-related options
are enabled in the BIOS): 

root@pcitfiot05:/usr/src/pmtools-20070714/acpidump# ./acpidump
ACPI tables were not found. If you know location of RSD PTR table (from
dmesg, etc), supply it with either --addr or -a option
root@pcitfiot05:/usr/src/pmtools-20070714/acpidump# dmesg | grep ACPI | head -20
 BIOS-e820: 00000000cff60000 - 00000000cff69000 (ACPI data)
 BIOS-e820: 00000000cff69000 - 00000000cff80000 (ACPI NVS)
ACPI: RSDP 000F6030, 0014 (r0 PTLTD )
ACPI: RSDT CFF63C39, 003C (r1 PTLTD    RSDT    6040000  LTP        0)
ACPI: FACP CFF68E10, 0074 (r1 INTEL  TUMWATER  6040000 PTL         3)
ACPI: DSDT CFF65C59, 31B7 (r1  Intel BLAKFORD  6040000 MSFT  100000E)
ACPI: FACS CFF69FC0, 0040
ACPI: APIC CFF68E84, 00C8 (r1 PTLTD      APIC    6040000  LTP        0)
ACPI: MCFG CFF68F4C, 003C (r1 PTLTD    MCFG    6040000  LTP        0)
ACPI: BOOT CFF68F88, 0028 (r1 PTLTD  $SBFTBL$  6040000  LTP        1)
ACPI: SPCR CFF68FB0, 0050 (r1 PTLTD  $UCRTBL$  6040000 PTL         1)
ACPI: SSDT CFF63C75, 1405 (r1  PmRef    CpuPm     3000 INTL 20050228)
ACPI: PM-Timer IO Port: 0x1008
ACPI: Local APIC address 0xfee00000
ACPI: LAPIC (acpi_id[0x00] lapic_id[0x00] enabled)
ACPI: LAPIC (acpi_id[0x01] lapic_id[0x04] enabled)
ACPI: LAPIC (acpi_id[0x02] lapic_id[0x01] enabled)
ACPI: LAPIC (acpi_id[0x03] lapic_id[0x05] enabled)
ACPI: LAPIC (acpi_id[0x04] lapic_id[0x02] enabled)
ACPI: LAPIC (acpi_id[0x05] lapic_id[0x06] enabled)

This is with the latest BIOS (Supermicro X7DBT, 1.3c); it has an option
called "Processor Power Management" which can be set to either 

"GV1/GV3 only" - according to the BIOS, this enables GV1/GV3 (whatever
that is) but no C states. This is what I use.
"Disabled" - which disables everything.

According to the BIOS help there should be two more settings to that
that give me C states but those are only described, they don't appear in
the selection box. 

Two more options called "Thermal Management 2" and "C1 Enhanced Mode"
were also enabled, still no C states.

I guess I'll be nagging Supermicro about this.

thanks again,
Andras

Re: Intel Clovertown vs. frequency scaling

[Andi Kleen]

Andras.Horvath@cern.ch writes:
>
> According to the BIOS help there should be two more settings to that
> that give me C states but those are only described, they don't appear in
> the selection box. 

Desktop/Server boxes normally don't have any deeper C states than C1
and C1 is just the traditional "HLT" instruction and doesn't need
ACPI support at all and is usually not described in ACPI tables
when it's the only state.

The BIOS options likely just control what the BIOS' SMM handler does
hidden inside C1.

-Andi