[PATCHSET] 0/3 A dynamic cpufreq governor

[PATCHSET] 0/3 A dynamic cpufreq governor

[Pallipadi, Venkatesh]

Hi,

This is the "new and improved" version of the patch  
that was sent out earlier this week. Thanks for all the 
feedback/suggestions.
(lkml subject line reference for previous thread -
"Dynamic cpufreq governor and updates to ACPI P-state driver")

The changes from the previous version:
1) dynamic cpufreq governor changes completely separated 
  from the updates to acpi P-state driver. These patches only 
  deal with dynamic cpufreq governor. Separate patch will be 
  sent out later to cover acpi updates.
2) New name for this cpufreq driver "cpufreq_ondemand"
  (Thanks to Pavel for the name)
3) Miscellaneous bugfixes and changes based on the review comments
4) And introduction of tuning knobs, where advanced user can tune
  the cpufreq_ondemand behaviour. 
  (Thanks to Dominik for helping me with this)
  Sampling_rate, down_threshold, up_threshold are currently exported.
  More advanced user who wants to change the algorithm itself, can 
  take the base infrastructure from this driver and come up with a 
  new/optimized frequency switching algorithm, as this driver is GPLed
:).

Patch contents (patches against 2.6.0-test8):
ondemand1.patch - 'unit' related bugfixes in drivers. 

ondemand2.patch - basic cpufreq_ondemand governor, with some changes 
since the last time.

ondemand3.patch - Adding sysfs interface for cpufreq_ondemand 
tunables and making sysfs access by cpufreq governors safe against 
removal of the underlying module (from Dominik). 


Thanks,
-Venkatesh

[PATCHSET] 0/3 A dynamic cpufreq governor

[Pallipadi, Venkatesh]

Hi,

This is the "new and improved" version of the patch  
that was sent out earlier this week. Thanks for all the 
feedback/suggestions.
(lkml subject line reference for previous thread -
"Dynamic cpufreq governor and updates to ACPI P-state driver")

The changes from the previous version:
1) dynamic cpufreq governor changes completely separated 
  from the updates to acpi P-state driver. These patches only 
  deal with dynamic cpufreq governor. Separate patch will be 
  sent out later to cover acpi updates.
2) New name for this cpufreq driver "cpufreq_ondemand"
  (Thanks to Pavel for the name)
3) Miscellaneous bugfixes and changes based on the review comments
4) And introduction of tuning knobs, where advanced user can tune
  the cpufreq_ondemand behaviour. 
  (Thanks to Dominik for helping me with this)
  Sampling_rate, down_threshold, up_threshold are currently exported.
  More advanced user who wants to change the algorithm itself, can 
  take the base infrastructure from this driver and come up with a 
  new/optimized frequency switching algorithm, as this driver is GPLed
:).

Patch contents (patches against 2.6.0-test8):
ondemand1.patch - 'unit' related bugfixes in drivers. 

ondemand2.patch - basic cpufreq_ondemand governor, with some changes 
since the last time.

ondemand3.patch - Adding sysfs interface for cpufreq_ondemand 
tunables and making sysfs access by cpufreq governors safe against 
removal of the underlying module (from Dominik). 


Thanks,
-Venkatesh

Re: [PATCHSET] 0/3 A dynamic cpufreq governor

[Måns Rullgård]

"Pallipadi, Venkatesh" <venkatesh.pallipadi@intel.com> writes:

> This is the "new and improved" version of the patch  
> that was sent out earlier this week. Thanks for all the 
> feedback/suggestions.
> (lkml subject line reference for previous thread -
> "Dynamic cpufreq governor and updates to ACPI P-state driver")

Is there any way this thing can be made to work on a Pentium 4 M,
using p4-clockmod for frequency changing?

-- 
Måns Rullgård
mru@kth.se

Re: [PATCHSET] 0/3 A dynamic cpufreq governor

[Måns Rullgård]

"Pallipadi, Venkatesh" <venkatesh.pallipadi@intel.com> writes:

> This is the "new and improved" version of the patch  
> that was sent out earlier this week. Thanks for all the 
> feedback/suggestions.
> (lkml subject line reference for previous thread -
> "Dynamic cpufreq governor and updates to ACPI P-state driver")

Is there any way this thing can be made to work on a Pentium 4 M,
using p4-clockmod for frequency changing?

-- 
Måns Rullgård
mru@kth.se

Re: [PATCHSET] 0/3 A dynamic cpufreq governor

[Dominik Brodowski]

On Sat, Oct 25, 2003 at 10:07:59PM +0200, Måns Rullgård wrote:
> "Pallipadi, Venkatesh" <venkatesh.pallipadi@intel.com> writes:
> 
> > This is the "new and improved" version of the patch  
> > that was sent out earlier this week. Thanks for all the 
> > feedback/suggestions.
> > (lkml subject line reference for previous thread -
> > "Dynamic cpufreq governor and updates to ACPI P-state driver")
> 
> Is there any way this thing can be made to work on a Pentium 4 M,
> using p4-clockmod for frequency changing?

You should prefer the speedstep-ich driver for Pentium 4 M, as long as the
southbridge on your notebook is from Intel - p4-clockmod doesn't save you
anything if either APM or ACPI "idling" works correctly.

	Dominik

Re: [PATCHSET] 0/3 A dynamic cpufreq governor

[Måns Rullgård]

Dominik Brodowski <linux@brodo.de> writes:

>> > This is the "new and improved" version of the patch  
>> > that was sent out earlier this week. Thanks for all the 
>> > feedback/suggestions.
>> > (lkml subject line reference for previous thread -
>> > "Dynamic cpufreq governor and updates to ACPI P-state driver")
>> 
>> Is there any way this thing can be made to work on a Pentium 4 M,
>> using p4-clockmod for frequency changing?
>
> You should prefer the speedstep-ich driver for Pentium 4 M, as long as the
> southbridge on your notebook is from Intel - p4-clockmod doesn't save you

The problem is my chipset is SiS 650.

> anything if either APM or ACPI "idling" works correctly.

-- 
Måns Rullgård
mru@kth.se

Re: [PATCHSET] 0/3 A dynamic cpufreq governor

[Ducrot Bruno]

On Mon, Oct 27, 2003 at 04:23:22PM +0100, Måns Rullgård wrote:
> Dominik Brodowski <linux@brodo.de> writes:
> 
> >> > This is the "new and improved" version of the patch  
> >> > that was sent out earlier this week. Thanks for all the 
> >> > feedback/suggestions.
> >> > (lkml subject line reference for previous thread -
> >> > "Dynamic cpufreq governor and updates to ACPI P-state driver")
> >> 
> >> Is there any way this thing can be made to work on a Pentium 4 M,
> >> using p4-clockmod for frequency changing?
> >
> > You should prefer the speedstep-ich driver for Pentium 4 M, as long as the
> > southbridge on your notebook is from Intel - p4-clockmod doesn't save you
> 
> The problem is my chipset is SiS 650.
> 

Does the generic acpi P-state driver work?  If not, well, nobody can't
help you, if no one work for the reverse engeenering part for SiS.
It was so long for the PIIX4 case (but probably the SiS should be a little
bit more easy like the ich case I hope).

-- 
Ducrot Bruno

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

Re: [PATCHSET] 0/3 A dynamic cpufreq governor

[Måns Rullgård]

Ducrot Bruno <ducrot@poupinou.org> writes:

>> >> Is there any way this thing can be made to work on a Pentium 4 M,
>> >> using p4-clockmod for frequency changing?
>> >
>> > You should prefer the speedstep-ich driver for Pentium 4 M, as long as the
>> > southbridge on your notebook is from Intel - p4-clockmod doesn't save you
>> 
>> The problem is my chipset is SiS 650.
>> 
>
> Does the generic acpi P-state driver work?  If not, well, nobody can't

I think so, but it only gives two different speeds to choose from,
1200 MHz and 2 GHz.  p4-clockmod give me speeds from 250 MHz to 2 GHz.

-- 
Måns Rullgård
mru@kth.se

Re: [PATCHSET] 0/3 A dynamic cpufreq governor

[Ducrot Bruno]

On Mon, Oct 27, 2003 at 05:06:31PM +0100, Måns Rullgård wrote:
> Ducrot Bruno <ducrot@poupinou.org> writes:
> 
> >> >> Is there any way this thing can be made to work on a Pentium 4 M,
> >> >> using p4-clockmod for frequency changing?
> >> >
> >> > You should prefer the speedstep-ich driver for Pentium 4 M, as long as the
> >> > southbridge on your notebook is from Intel - p4-clockmod doesn't save you
> >> 
> >> The problem is my chipset is SiS 650.
> >> 
> >
> > Does the generic acpi P-state driver work?  If not, well, nobody can't
> 
> I think so, but it only gives two different speeds to choose from,
> 1200 MHz and 2 GHz.  p4-clockmod give me speeds from 250 MHz to 2 GHz.
> 

You have then voltage scaling with the acpi driver.
From my point-of-view, this is better than the p4-clockmod,
the speedstep-ich would give you no more states (if
it can be made to work with SiS southbridges), just that the acpi driver
will be a little bit slower (it work by calling some bios functions)
than direct access to IO like speedstep-ich does.
Note that p4-clockmod do *not* regulate voltage.  From power comsuptions
like, if you are at 1200 MHz, but via the acpi driver, you would be
like at 600MHz with the p4-clockmod, and actually it may be possible
that idle loops will be more power conservative with acpi (I am not sure for
that one though).  Dominik have better words than me to explain that.

-- 
Ducrot Bruno

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

Re: [PATCHSET] 0/3 A dynamic cpufreq governor

[Dominik Brodowski]

On Mon, Oct 27, 2003 at 05:19:05PM +0100, Ducrot Bruno wrote:
> On Mon, Oct 27, 2003 at 05:06:31PM +0100, Måns Rullgård wrote:
> > Ducrot Bruno <ducrot@poupinou.org> writes:
> > 
> > >> >> Is there any way this thing can be made to work on a Pentium 4 M,
> > >> >> using p4-clockmod for frequency changing?
> > >> >
> > >> > You should prefer the speedstep-ich driver for Pentium 4 M, as long as the
> > >> > southbridge on your notebook is from Intel - p4-clockmod doesn't save you
> > >> 
> > >> The problem is my chipset is SiS 650.
> > >> 
> > >
> > > Does the generic acpi P-state driver work?  If not, well, nobody can't
> > 
> > I think so, but it only gives two different speeds to choose from,
> > 1200 MHz and 2 GHz.  p4-clockmod give me speeds from 250 MHz to 2 GHz.
> > 
> 
> You have then voltage scaling with the acpi driver.
> >From my point-of-view, this is better than the p4-clockmod,
> the speedstep-ich would give you no more states (if
> it can be made to work with SiS southbridges), just that the acpi driver
> will be a little bit slower (it work by calling some bios functions)
> than direct access to IO like speedstep-ich does.
> Note that p4-clockmod do *not* regulate voltage.  From power comsuptions
> like, if you are at 1200 MHz, but via the acpi driver, you would be
> like at 600MHz with the p4-clockmod, and actually it may be possible
> that idle loops will be more power conservative with acpi (I am not sure for
> that one though).  Dominik have better words than me to explain that.

For more details and calculations, check this thread:

http://www.linux.org.uk/mailman/private/cpufreq/2003-September/002495.html


	Dominik