hawkboard at 456 Mhz?

hawkboard at 456 Mhz?

[Frans Meulenbroeks]

Hi,

In the openembedded repository I saw the following patch:
http://cgit.openembedded.org/cgit.cgi/openembedded/tree/recipes/linux/linux-davinci/da850_omapl138_opp456mhz.patch
This seems to suggest that it is possible to run the hawkboard at 456
Mhz (and yeah: I know it is out of spec and out of warranty, but for
an experiment it would be quite helpful for me).
I've build the kernel with openembedded, the patch is applied,
CONF_CPU_FREQ is defined, but the control hook in /sys is not there.
I've also tried tweaking the file with the settings (ds850.c) but to no avail.

Anyone a suggestion on how to get this done?

Thanks alot! FM

PS: a way to downclock would also be fine for my experiment. For a
disk benchmark I would like to know what part of the time it takes to
read data actually is caused by the device and which comes from the
os.
Assumption is that the device time is independent of the CPU clock and
that the os time is roughly linear with the cpu frequency. (and yes, I
use an access pattern that tries to avoid pre-caching by the device).

Re: hawkboard at 456 Mhz?

[Koen Kooi]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 21-03-10 13:57, Frans Meulenbroeks wrote:
> Hi,
> 
> In the openembedded repository I saw the following patch:
> http://cgit.openembedded.org/cgit.cgi/openembedded/tree/recipes/linux/linux-davinci/da850_omapl138_opp456mhz.patch
> This seems to suggest that it is possible to run the hawkboard at 456
> Mhz (and yeah: I know it is out of spec and out of warranty, but for
> an experiment it would be quite helpful for me).
> I've build the kernel with openembedded, the patch is applied,
> CONF_CPU_FREQ is defined, but the control hook in /sys is not there.
> I've also tried tweaking the file with the settings (ds850.c) but to no avail.
> 
> Anyone a suggestion on how to get this done?

You need a newer kernel that doesn't rely on the TPS part for cpufreq
(hawkboard lacks a PMIC, the evm uses the TPS). You can build from the
ti/staging[1] branch to get such a kernel.

Or wait a week for that branch to get merge, internal testing is finally
wrapping up.

[1] http://gitorious.org/angstrom/openembedded/commits/ti/staging
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.5 (Darwin)

iD8DBQFLpjBLMkyGM64RGpERAs/NAJ9OK1jN2jxYYj12gWKu2vxCrNsCMACfelgj
+b0keOpVlJBB+WYeL4fJlyg=
=5LGb
-----END PGP SIGNATURE-----

Re: hawkboard at 456 Mhz?

[Frans Meulenbroeks]

2010/3/21 Koen Kooi <k.kooi@student.utwente.nl>:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> On 21-03-10 13:57, Frans Meulenbroeks wrote:
>> Hi,
>>
>> In the openembedded repository I saw the following patch:
>> http://cgit.openembedded.org/cgit.cgi/openembedded/tree/recipes/linux/linux-davinci/da850_omapl138_opp456mhz.patch
>> This seems to suggest that it is possible to run the hawkboard at 456
>> Mhz (and yeah: I know it is out of spec and out of warranty, but for
>> an experiment it would be quite helpful for me).
>> I've build the kernel with openembedded, the patch is applied,
>> CONF_CPU_FREQ is defined, but the control hook in /sys is not there.
>> I've also tried tweaking the file with the settings (ds850.c) but to no avail.
>>
>> Anyone a suggestion on how to get this done?
>
> You need a newer kernel that doesn't rely on the TPS part for cpufreq
> (hawkboard lacks a PMIC, the evm uses the TPS). You can build from the
> ti/staging[1] branch to get such a kernel.
>
> Or wait a week for that branch to get merge, internal testing is finally
> wrapping up.
>
> [1] http://gitorious.org/angstrom/openembedded/commits/ti/staging


Koen,

Thanks alot!
As patience is something that I don't have an overdose of, I decided
not to waste any on this issue.
So I checked out the git and rebuild the kernel. Works like a charm.

For those who want to reproduce this:
Be sure to adapt defconfig (CONFIG_DA8XX_MAX_SPEED_456=y)

I changed the speed setting with cpuspeed, this one is not on the
angstrom feed, so you need to build it yourself too.

(I have assumed that you can rebuild an openembedded kernel, if
not,then probably this is not for you).

Havve fun!
Frans.

Re: hawkboard at 456 Mhz?

[Koen Kooi]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 21-03-10 20:38, Frans Meulenbroeks wrote:
> 2010/3/21 Koen Kooi <k.kooi@student.utwente.nl>:
>> -----BEGIN PGP SIGNED MESSAGE-----
>> Hash: SHA1
>>
>> On 21-03-10 13:57, Frans Meulenbroeks wrote:
>>> Hi,
>>>
>>> In the openembedded repository I saw the following patch:
>>> http://cgit.openembedded.org/cgit.cgi/openembedded/tree/recipes/linux/linux-davinci/da850_omapl138_opp456mhz.patch
>>> This seems to suggest that it is possible to run the hawkboard at 456
>>> Mhz (and yeah: I know it is out of spec and out of warranty, but for
>>> an experiment it would be quite helpful for me).
>>> I've build the kernel with openembedded, the patch is applied,
>>> CONF_CPU_FREQ is defined, but the control hook in /sys is not there.
>>> I've also tried tweaking the file with the settings (ds850.c) but to no avail.
>>>
>>> Anyone a suggestion on how to get this done?
>>
>> You need a newer kernel that doesn't rely on the TPS part for cpufreq
>> (hawkboard lacks a PMIC, the evm uses the TPS). You can build from the
>> ti/staging[1] branch to get such a kernel.
>>
>> Or wait a week for that branch to get merge, internal testing is finally
>> wrapping up.
>>
>> [1] http://gitorious.org/angstrom/openembedded/commits/ti/staging
> 
> 
> Koen,
> 
> Thanks alot!
> As patience is something that I don't have an overdose of, I decided
> not to waste any on this issue.
> So I checked out the git and rebuild the kernel. Works like a charm.
> 
> For those who want to reproduce this:
> Be sure to adapt defconfig (CONFIG_DA8XX_MAX_SPEED_456=y)
> 
> I changed the speed setting with cpuspeed, this one is not on the
> angstrom feed, so you need to build it yourself too.

I use cpufrequtils, is cpuspeed a better option?
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.5 (Darwin)

iD8DBQFLpyu/MkyGM64RGpERAp9qAJ4ppZ8U3HCBTD+zp+TUs2Y6+vNeAwCdElnj
0Lk6h8fnSQjG2G58ihbMy20=
=6PRK
-----END PGP SIGNATURE-----

Re: hawkboard at 456 Mhz?

[Frans Meulenbroeks]

2010/3/22 Koen Kooi <k.kooi@student.utwente.nl>:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> On 21-03-10 20:38, Frans Meulenbroeks wrote:
>> 2010/3/21 Koen Kooi <k.kooi@student.utwente.nl>:
>>> -----BEGIN PGP SIGNED MESSAGE-----
>>> Hash: SHA1
>>>
>>> On 21-03-10 13:57, Frans Meulenbroeks wrote:
>>>> Hi,
>>>>
>>>> In the openembedded repository I saw the following patch:
>>>> http://cgit.openembedded.org/cgit.cgi/openembedded/tree/recipes/linux/linux-davinci/da850_omapl138_opp456mhz.patch
>>>> This seems to suggest that it is possible to run the hawkboard at 456
>>>> Mhz (and yeah: I know it is out of spec and out of warranty, but for
>>>> an experiment it would be quite helpful for me).
>>>> I've build the kernel with openembedded, the patch is applied,
>>>> CONF_CPU_FREQ is defined, but the control hook in /sys is not there.
>>>> I've also tried tweaking the file with the settings (ds850.c) but to no avail.
>>>>
>>>> Anyone a suggestion on how to get this done?
>>>
>>> You need a newer kernel that doesn't rely on the TPS part for cpufreq
>>> (hawkboard lacks a PMIC, the evm uses the TPS). You can build from the
>>> ti/staging[1] branch to get such a kernel.
>>>
>>> Or wait a week for that branch to get merge, internal testing is finally
>>> wrapping up.
>>>
>>> [1] http://gitorious.org/angstrom/openembedded/commits/ti/staging
>>
>>
>> Koen,
>>
>> Thanks alot!
>> As patience is something that I don't have an overdose of, I decided
>> not to waste any on this issue.
>> So I checked out the git and rebuild the kernel. Works like a charm.
>>
>> For those who want to reproduce this:
>> Be sure to adapt defconfig (CONFIG_DA8XX_MAX_SPEED_456=y)
>>
>> I changed the speed setting with cpuspeed, this one is not on the
>> angstrom feed, so you need to build it yourself too.
>
> I use cpufrequtils, is cpuspeed a better option?

I was unaware of cpufrequtils and bumped into cpuspeed, so I used that.
I just quickly browsed the manpage of cpufreq-set and it would have
served me better (as cupfreq-set has a -f option to force the speed
and cpuspeed is controlled to signals.
Here is the manpage for cpuspeed: http://linux.die.net/man/8/cpuspeed

Btw I would have expected that I could also directly poke in /sys to
change the speed but the one attempt I did gave an error. Might be
that I just used the wrong var
(actually i probably had to use scaling_set_speed and I probably used
something like actual_speed or current_speed).

Frans.

PS: i've briefly peeked at the processor doc: would it be possible to
push things even further (e.g. to 600 Mhz). Of course I know this is
out of warranty, but that is 456 too :-) I'm still missing some info
on how to play with what to do an educated attempt (it is manly the
voltage related settings I feel unsure about).

Re: hawkboard at 456 Mhz?

[Frans Meulenbroeks]

2010/3/22 Frans Meulenbroeks <fransmeulenbroeks@gmail.com>:
> 2010/3/22 Koen Kooi <k.kooi@student.utwente.nl>:
>> -----BEGIN PGP SIGNED MESSAGE-----
>> Hash: SHA1
>>
>> On 21-03-10 20:38, Frans Meulenbroeks wrote:
>>> 2010/3/21 Koen Kooi <k.kooi@student.utwente.nl>:
>>>> -----BEGIN PGP SIGNED MESSAGE-----
>>>> Hash: SHA1
>>>>
>>>> On 21-03-10 13:57, Frans Meulenbroeks wrote:
>>>>> Hi,
>>>>>
>>>>> In the openembedded repository I saw the following patch:
>>>>> http://cgit.openembedded.org/cgit.cgi/openembedded/tree/recipes/linux/linux-davinci/da850_omapl138_opp456mhz.patch
>>>>> This seems to suggest that it is possible to run the hawkboard at 456
>>>>> Mhz (and yeah: I know it is out of spec and out of warranty, but for
>>>>> an experiment it would be quite helpful for me).
>>>>> I've build the kernel with openembedded, the patch is applied,
>>>>> CONF_CPU_FREQ is defined, but the control hook in /sys is not there.
>>>>> I've also tried tweaking the file with the settings (ds850.c) but to no avail.
>>>>>
>>>>> Anyone a suggestion on how to get this done?
>>>>
>>>> You need a newer kernel that doesn't rely on the TPS part for cpufreq
>>>> (hawkboard lacks a PMIC, the evm uses the TPS). You can build from the
>>>> ti/staging[1] branch to get such a kernel.
>>>>
>>>> Or wait a week for that branch to get merge, internal testing is finally
>>>> wrapping up.
>>>>
>>>> [1] http://gitorious.org/angstrom/openembedded/commits/ti/staging
>>>
>>>
>>> Koen,
>>>
>>> Thanks alot!
>>> As patience is something that I don't have an overdose of, I decided
>>> not to waste any on this issue.
>>> So I checked out the git and rebuild the kernel. Works like a charm.
>>>
>>> For those who want to reproduce this:
>>> Be sure to adapt defconfig (CONFIG_DA8XX_MAX_SPEED_456=y)
>>>
>>> I changed the speed setting with cpuspeed, this one is not on the
>>> angstrom feed, so you need to build it yourself too.
>>
>> I use cpufrequtils, is cpuspeed a better option?
>
> I was unaware of cpufrequtils and bumped into cpuspeed, so I used that.
> I just quickly browsed the manpage of cpufreq-set and it would have
> served me better (as cupfreq-set has a -f option to force the speed
> and cpuspeed is controlled to signals.
> Here is the manpage for cpuspeed: http://linux.die.net/man/8/cpuspeed
>
> Btw I would have expected that I could also directly poke in /sys to
> change the speed but the one attempt I did gave an error. Might be
> that I just used the wrong var
> (actually i probably had to use scaling_set_speed and I probably used
> something like actual_speed or current_speed).
>
> Frans.
>
> PS: i've briefly peeked at the processor doc: would it be possible to
> push things even further (e.g. to 600 Mhz). Of course I know this is
> out of warranty, but that is 456 too :-) I'm still missing some info
> on how to play with what to do an educated attempt (it is manly the
> voltage related settings I feel unsure about).
>
Forgot to mention: google yesterday gave me also a gtk variant:
gtk-cpuspeedy and a kde one: klaptopdaemon
didn't try these.

FM