* Re: Cpu frequency scaling [not found] <200301101600.26246.krishnakumar@naturesoft.net> @ 2003-01-12 17:09 ` Ralf Baechle 2003-01-12 19:01 ` Kevin D. Kissell 0 siblings, 1 reply; 12+ messages in thread From: Ralf Baechle @ 2003-01-12 17:09 UTC (permalink / raw) To: Krishnakumar. R; +Cc: linux-mips On Sun, Jan 12, 2003 at 08:20:05PM +0530, Krishnakumar. R wrote: > Can frequency scaling (through software) > be done on mips using linux ? > > Is such a feature feasible in mips ?? > > I could not find any documentation nor patches > for frequency > scaling on mips at > http://www.brodo.de/cpufreq/ > :-( None of the currently supported MIPS CPUs support such a feature in hardware as such our support is already complete and by definition bug free ;-) Ralf ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: Cpu frequency scaling @ 2003-01-12 19:01 ` Kevin D. Kissell 0 siblings, 0 replies; 12+ messages in thread From: Kevin D. Kissell @ 2003-01-12 19:01 UTC (permalink / raw) To: Ralf Baechle, Krishnakumar. R; +Cc: linux-mips > > Can frequency scaling (through software) > > be done on mips using linux ? > > > > Is such a feature feasible in mips ?? > > > > I could not find any documentation nor patches > > for frequency > > scaling on mips at > > http://www.brodo.de/cpufreq/ > > :-( > > None of the currently supported MIPS CPUs support such a feature in > hardware as such our support is already complete and by definition > bug free ;-) Actually, that's not *quite* true. A number of MIPS CPUs and cores that are otherwise supported by Linux (e.g. 4K, 5K) have a "reduced power" mode which is modulated by the CP0 "RP" bit. In general, this bit does nothing in the CPU itself, however. It was intended that it be connected to system-level logic for frequency scaling (1/n normal clock, or CPU=Bus). So on most systems it does nothing, and on the ones where it does do something, it's entirely system dependent. I don't have an Alchemy AU1000 spec handy, but since they've integrated a lot of other logic with their CPU, and since they designed their component to go into low-power devices, it wouldn't surprise me in the least if they do something well-defined with the RP bit. So, to get back to the original question, something highly platform dependent *could* be done using MIPS/Linux, via /proc/cpu or some kind of system call, but I don't believe anyone has made such a hook generally available as yet. Kevin K. ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: Cpu frequency scaling @ 2003-01-12 19:01 ` Kevin D. Kissell 0 siblings, 0 replies; 12+ messages in thread From: Kevin D. Kissell @ 2003-01-12 19:01 UTC (permalink / raw) To: Ralf Baechle, Krishnakumar. R; +Cc: linux-mips > > Can frequency scaling (through software) > > be done on mips using linux ? > > > > Is such a feature feasible in mips ?? > > > > I could not find any documentation nor patches > > for frequency > > scaling on mips at > > http://www.brodo.de/cpufreq/ > > :-( > > None of the currently supported MIPS CPUs support such a feature in > hardware as such our support is already complete and by definition > bug free ;-) Actually, that's not *quite* true. A number of MIPS CPUs and cores that are otherwise supported by Linux (e.g. 4K, 5K) have a "reduced power" mode which is modulated by the CP0 "RP" bit. In general, this bit does nothing in the CPU itself, however. It was intended that it be connected to system-level logic for frequency scaling (1/n normal clock, or CPU=Bus). So on most systems it does nothing, and on the ones where it does do something, it's entirely system dependent. I don't have an Alchemy AU1000 spec handy, but since they've integrated a lot of other logic with their CPU, and since they designed their component to go into low-power devices, it wouldn't surprise me in the least if they do something well-defined with the RP bit. So, to get back to the original question, something highly platform dependent *could* be done using MIPS/Linux, via /proc/cpu or some kind of system call, but I don't believe anyone has made such a hook generally available as yet. Kevin K. ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: Cpu frequency scaling 2003-01-12 19:01 ` Kevin D. Kissell (?) @ 2003-01-13 4:15 ` Pete Popov -1 siblings, 0 replies; 12+ messages in thread From: Pete Popov @ 2003-01-13 4:15 UTC (permalink / raw) To: Kevin D. Kissell; +Cc: Ralf Baechle, Krishnakumar. R, linux-mips On Sun, 2003-01-12 at 11:01, Kevin D. Kissell wrote: > > > Can frequency scaling (through software) > > > be done on mips using linux ? > > > > > > Is such a feature feasible in mips ?? > > > > > > I could not find any documentation nor patches > > > for frequency > > > scaling on mips at > > > http://www.brodo.de/cpufreq/ > > > :-( > > > > None of the currently supported MIPS CPUs support such a feature in > > hardware as such our support is already complete and by definition > > bug free ;-) > > Actually, that's not *quite* true. A number of MIPS CPUs and > cores that are otherwise supported by Linux (e.g. 4K, 5K) have > a "reduced power" mode which is modulated by the CP0 "RP" > bit. In general, this bit does nothing in the CPU itself, however. > It was intended that it be connected to system-level logic for > frequency scaling (1/n normal clock, or CPU=Bus). So on > most systems it does nothing, and on the ones where it does > do something, it's entirely system dependent. I don't have an > Alchemy AU1000 spec handy, but since they've integrated > a lot of other logic with their CPU, and since they designed > their component to go into low-power devices, it wouldn't > surprise me in the least if they do something well-defined > with the RP bit. Actually the Au1x CPUs support both, frequency and voltage scaling. > So, to get back to the original question, something highly > platform dependent *could* be done using MIPS/Linux, > via /proc/cpu or some kind of system call, but I don't believe > anyone has made such a hook generally available as yet. For the Au1x CPUs, I had added a /proc interface that allows you to do something like "echo 192 > /proc/cpufreq" to reduce the frequency from 396 MHz to 192, or whatever number you chose, but the implementation was very adhoc and probably not generally useful for other CPUs. IBM and MontaVista recently published a whitepaper on power management that's designed to be arch independent. The design considers not only CPU scaling but scaling of all the buses as well, which I don't think the cpufreq project takes into account. Pete ^ permalink raw reply [flat|nested] 12+ messages in thread
* CPU Frequency Scaling
@ 2004-09-01 23:01 ncrfgs
2004-09-02 9:48 ` Bruno Ducrot
0 siblings, 1 reply; 12+ messages in thread
From: ncrfgs @ 2004-09-01 23:01 UTC (permalink / raw)
To: cpufreq
[-- Attachment #1.1: Type: text/plain, Size: 2640 bytes --]
Hi,
I'm running a vanilla 2.6.8.1 Linux kernel on an iBook 2.2 G3
900 MHz and I'd like to try the CPU frequency scaling.
Is it required to enable "/proc/sys/cpu/ interface (2.4. / OLD)"
in order to be able to use CPU frequency scaling efectively?
The kernel I'm running doesn't include it, I'm using powernowd
as userspace governor and the CPU frequency scaling seems to
work as far as I know from what I read in /proc/cpu* and
/sys/devices/system/cpu/cpu0/cpufreq/* files.
Unfortunately even when /proc/cpuinfo reports a clock of 900MHz
and /sys/devices/system/cpu/cpu0/cpufreq/scaling_cur_freq is set
to 900000 the system looks quite slow to me, for example during
compilation processes, as if the current frequency was 400 MHz.
Do I have to enable "Support for Apple PowerBooks", too?
Some days ago I argued about this issue with a guy from the
debian-powerpc mailing list, too and I faced a strange situation
I didn't understand. According to his kernel config he is using
'performance' as Default CPUFreq governor but as far as I can see
from the 'performance' governor Help it shouldn't allow CPU
Frequency scaling:
CONFIG_CPU_FREQ_GOV_PERFORMANCE:
This cpufreq governors set the frequency statically to the
highest available CPU frequency.
He said that's strange indeed. Nevertheless, He's sure that his frequency is
changing...
"Because the battery life increase by 1 hour with cpufreqd. And Battery
load derivative decrease (absolute value increase) when I use some cpu
intensive programs. Less formally, the more cpu I use the faster my
battery load decrease."
Is it just an impression of his?
Here a portion of his kernel .config file follows.
CONFIG_CPU_FREQ=y
# CONFIG_CPU_FREQ_PROC_INTF is not set
CONFIG_CPU_FREQ_DEFAULT_GOV_PERFORMANCE=y
# CONFIG_CPU_FREQ_DEFAULT_GOV_USERSPACE is not set
CONFIG_CPU_FREQ_GOV_PERFORMANCE=y
CONFIG_CPU_FREQ_GOV_POWERSAVE=m
CONFIG_CPU_FREQ_GOV_USERSPACE=m
# CONFIG_CPU_FREQ_24_API is not set
CONFIG_CPU_FREQ_PMAC=y
CONFIG_CPU_FREQ_TABLE=y
As far as I could see from the section of the kernel config he sent me
last time, he compiled 'userspace' governor as a module. So I asked him
if was it loeaded but he answered that is was not.
> cat devices/system/cpu/cpu0/cpufreq/scaling_available_governors
> performance
Thanks in advance.
Best regards.
--
Value your freedom, or you will lose it, teaches history.
``Don't bother us with politics,'' respond those who don't
want to learn.
-- Richard M. Stallman
http://www.gnu.org/philosophy/linux-gnu-freedom.html
[-- Attachment #1.2: Type: application/pgp-signature, Size: 189 bytes --]
[-- Attachment #2: Type: text/plain, Size: 143 bytes --]
_______________________________________________
Cpufreq mailing list
Cpufreq@www.linux.org.uk
http://www.linux.org.uk/mailman/listinfo/cpufreq
^ permalink raw reply [flat|nested] 12+ messages in thread* Re: CPU Frequency Scaling 2004-09-01 23:01 CPU Frequency Scaling ncrfgs @ 2004-09-02 9:48 ` Bruno Ducrot 2004-09-02 10:53 ` Mattia Dongili 0 siblings, 1 reply; 12+ messages in thread From: Bruno Ducrot @ 2004-09-02 9:48 UTC (permalink / raw) To: ncrfgs; +Cc: cpufreq On Thu, Sep 02, 2004 at 01:01:03AM +0200, ncrfgs wrote: > Hi, > > I'm running a vanilla 2.6.8.1 Linux kernel on an iBook 2.2 G3 > 900 MHz and I'd like to try the CPU frequency scaling. > > Is it required to enable "/proc/sys/cpu/ interface (2.4. / OLD)" > in order to be able to use CPU frequency scaling efectively? No. It's old, somehow ugly, and will be removed in the future. > The kernel I'm running doesn't include it, I'm using powernowd > as userspace governor and the CPU frequency scaling seems to > work as far as I know from what I read in /proc/cpu* and > /sys/devices/system/cpu/cpu0/cpufreq/* files. > > Unfortunately even when /proc/cpuinfo reports a clock of 900MHz > and /sys/devices/system/cpu/cpu0/cpufreq/scaling_cur_freq is set > to 900000 the system looks quite slow to me, for example during > compilation processes, as if the current frequency was 400 MHz. > > Do I have to enable "Support for Apple PowerBooks", too? Yes, but for other reasons ;) (LCD brightness, suspend to ram, etc.) (I think the G3 don't use the pmu, though I guess benh know more than me for this..) > Some days ago I argued about this issue with a guy from the > debian-powerpc mailing list, too and I faced a strange situation > I didn't understand. According to his kernel config he is using > 'performance' as Default CPUFreq governor but as far as I can see > from the 'performance' governor Help it shouldn't allow CPU > Frequency scaling: > > CONFIG_CPU_FREQ_GOV_PERFORMANCE: > > This cpufreq governors set the frequency statically to the > highest available CPU frequency. No. For two reason: first: the iBook set on boot the processor in the lowest available frequency, and the firmware will not change this. If the kernel is compiled with the performance governor as default, one frequency change will be done in order to set the processor at the highest frequency available. Second, the performance governor could be used to change the frequency by setting the maximal frequency allowed for it (look at the scaling_max_freq, which is writable, the 'real' max being in the cpuinfo_max_freq). By changing this value, you will set a new frequency (if the cpufreq driver associated is of course of type 'fixed', not like the longrun, for example). > He said that's strange indeed. Nevertheless, He's sure that his frequency is > changing... Nope, nothing strange ;) > "Because the battery life increase by 1 hour with cpufreqd. And Battery > load derivative decrease (absolute value increase) when I use some cpu > intensive programs. Less formally, the more cpu I use the faster my > battery load decrease." > > Is it just an impression of his? He just forgot to check that the scaling_max_freq will change ;) PS: I don't know why this deamon use the performance governor instead of the userspace, or better why they don't use performance/powersave and tweak the (logical) min/max. This will ihmo make longrun owners happy... But since I don't use this deamon, perhaps it's already the case? -- Bruno Ducrot -- Which is worse: ignorance or apathy? -- Don't know. Don't care. ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: CPU Frequency Scaling 2004-09-02 9:48 ` Bruno Ducrot @ 2004-09-02 10:53 ` Mattia Dongili 2004-09-02 12:41 ` Bruno Ducrot 0 siblings, 1 reply; 12+ messages in thread From: Mattia Dongili @ 2004-09-02 10:53 UTC (permalink / raw) To: cpufreq On Thu, Sep 02, 2004 at 11:48:39AM +0200, Bruno Ducrot wrote: > On Thu, Sep 02, 2004 at 01:01:03AM +0200, ncrfgs wrote: [...] > > "Because the battery life increase by 1 hour with cpufreqd. And Battery > > load derivative decrease (absolute value increase) when I use some cpu > > intensive programs. Less formally, the more cpu I use the faster my > > battery load decrease." > > > > Is it just an impression of his? > > He just forgot to check that the scaling_max_freq will change ;) > > PS: I don't know why this deamon use the performance governor instead of the > userspace, or better why they don't use performance/powersave and tweak > the (logical) min/max. This will ihmo make longrun owners happy... But > since I don't use this deamon, perhaps it's already the case? Uhmmm... :) maybe I didn't understand your objection to cpufreqd, but let me spend a couple of words about it. Cpufreqd defines working Profiles as: [Profile] name=hi_boost minfreq=907060 maxfreq=997500 policy=performance where 'policy' can be any of the available kernel cpufreq governors. Anyway setting policy=userspace won't work as expected (I need to write it in the manpages...) because cpufreqd doesn't write to scaling_setspeed. The rationale behind that is that you can achive the same result as using the userspace governor by setting minfreq and maxfreq to the same value and use powersave or performance to round that value to the lower or higher value respectively. Or you can also achive the same by setting minfreq to the desired frequency and use the powersave governor with the same rounding capabilities (the same applies maxfreq/performance). The rationale behind the minfreq and maxfreq values is that I expect (one day) to find an 'on_demand' or similar governor that relies on scaling_min_freq and scaling_max_freq, not on values obtained by different sysfs entries. I'm trying to go in-depth enough to be able to write such a governor, but I need more time :) This simply follows the old Linus discussion on implementing policies in-kernel and give the user an higher level interface [1]. Does this (at least partially) answer your question? :) [1]: http://www.linux.org.uk/mailman/private/cpufreq/2002-August/000865.html -- mattia :wq! ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: CPU Frequency Scaling 2004-09-02 10:53 ` Mattia Dongili @ 2004-09-02 12:41 ` Bruno Ducrot 2004-09-02 13:36 ` Mattia Dongili 2004-09-03 8:34 ` ncrfgs 0 siblings, 2 replies; 12+ messages in thread From: Bruno Ducrot @ 2004-09-02 12:41 UTC (permalink / raw) To: cpufreq On Thu, Sep 02, 2004 at 12:53:48PM +0200, Mattia Dongili wrote: > Uhmmm... :) maybe I didn't understand your objection to cpufreqd, but > let me spend a couple of words about it. > > Cpufreqd defines working Profiles as: > [Profile] > name=hi_boost > minfreq=907060 > maxfreq=997500 > policy=performance > I know that already (and I have no objection since it's configurable via a configuration file and this can be changed for transmetta processors which is only my concern even though I don't have one). Anyway I supposed the OP was talking about powernowd, not cpufreqd, but maybe I was wrong ;) Cheers, -- Bruno Ducrot -- Which is worse: ignorance or apathy? -- Don't know. Don't care. ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: CPU Frequency Scaling 2004-09-02 12:41 ` Bruno Ducrot @ 2004-09-02 13:36 ` Mattia Dongili 2004-09-03 8:34 ` ncrfgs 1 sibling, 0 replies; 12+ messages in thread From: Mattia Dongili @ 2004-09-02 13:36 UTC (permalink / raw) To: cpufreq On Thu, Sep 02, 2004 at 02:41:51PM +0200, Bruno Ducrot wrote: > On Thu, Sep 02, 2004 at 12:53:48PM +0200, Mattia Dongili wrote: [...] > Anyway I supposed the OP was talking about powernowd, not cpufreqd, > but maybe I was wrong ;) Doh! :) I'm a little confused now but in the text I quoted from the OP you can find 'cpufreqd' cited :) and I thought you where referring to it. bye -- mattia :wq! ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: CPU Frequency Scaling 2004-09-02 12:41 ` Bruno Ducrot 2004-09-02 13:36 ` Mattia Dongili @ 2004-09-03 8:34 ` ncrfgs 2004-09-03 9:17 ` Bruno Ducrot 1 sibling, 1 reply; 12+ messages in thread From: ncrfgs @ 2004-09-03 8:34 UTC (permalink / raw) To: cpufreq; +Cc: Bruno Ducrot [-- Attachment #1.1: Type: text/plain, Size: 805 bytes --] On Thu, Sep 02, 2004 at 02:41:51PM +0200, Bruno Ducrot wrote: > Anyway I supposed the OP was talking about powernowd, not cpufreqd, > but maybe I was wrong ;) Actually I talked about both of them. =) Because the userspace governor I'm using is powernowd while the one used by the guy I talked with about CPU Frequency Scaling is cpufreqd. BTW which one would you suggest us to use? At the moment powernowd seems to work for me, but maybe you are aware of some problems or some issues I'm not. The same for cpufreqd. Thanks in advance. Best regards. -- Value your freedom, or you will lose it, teaches history. ``Don't bother us with politics,'' respond those who don't want to learn. -- Richard M. Stallman http://www.gnu.org/philosophy/linux-gnu-freedom.html [-- Attachment #1.2: Type: application/pgp-signature, Size: 189 bytes --] [-- Attachment #2: Type: text/plain, Size: 143 bytes --] _______________________________________________ Cpufreq mailing list Cpufreq@www.linux.org.uk http://www.linux.org.uk/mailman/listinfo/cpufreq ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: CPU Frequency Scaling 2004-09-03 8:34 ` ncrfgs @ 2004-09-03 9:17 ` Bruno Ducrot 2004-09-03 13:54 ` Carl Thompson 0 siblings, 1 reply; 12+ messages in thread From: Bruno Ducrot @ 2004-09-03 9:17 UTC (permalink / raw) To: cpufreq On Fri, Sep 03, 2004 at 10:34:50AM +0200, ncrfgs wrote: > BTW which one would you suggest us to use? > Maybe Carl, Mattia, Jeremy (and probably others) will answer ? :) Cheers, -- Bruno Ducrot -- Which is worse: ignorance or apathy? -- Don't know. Don't care. ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: CPU Frequency Scaling 2004-09-03 9:17 ` Bruno Ducrot @ 2004-09-03 13:54 ` Carl Thompson 0 siblings, 0 replies; 12+ messages in thread From: Carl Thompson @ 2004-09-03 13:54 UTC (permalink / raw) To: Bruno Ducrot; +Cc: cpufreq I've been meaning to release a new version of CPUSpeed (which is of course the best). I'll will definitely have time to do that over the long weekend. Check the CPUSpeed web site <http://carlthompson.net/software/cpuspeed> often! BTW I just realized that many of you that are using cpufreqd on recent versions of SuSE Linux are actually using my software. It seems the folks at SuSE renamed it. That is unfortunate because there is a different project that uses the cpufreqd name. Thanks, Carl Thompson Bruno Ducrot wrote: >On Fri, Sep 03, 2004 at 10:34:50AM +0200, ncrfgs wrote: > > >>BTW which one would you suggest us to use? >> >> >> > >Maybe Carl, Mattia, Jeremy (and probably others) will answer ? :) > >Cheers, > > > ^ permalink raw reply [flat|nested] 12+ messages in thread
end of thread, other threads:[~2004-09-03 13:54 UTC | newest]
Thread overview: 12+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
[not found] <200301101600.26246.krishnakumar@naturesoft.net>
2003-01-12 17:09 ` Cpu frequency scaling Ralf Baechle
2003-01-12 19:01 ` Kevin D. Kissell
2003-01-12 19:01 ` Kevin D. Kissell
2003-01-13 4:15 ` Pete Popov
2004-09-01 23:01 CPU Frequency Scaling ncrfgs
2004-09-02 9:48 ` Bruno Ducrot
2004-09-02 10:53 ` Mattia Dongili
2004-09-02 12:41 ` Bruno Ducrot
2004-09-02 13:36 ` Mattia Dongili
2004-09-03 8:34 ` ncrfgs
2004-09-03 9:17 ` Bruno Ducrot
2004-09-03 13:54 ` Carl Thompson
This is an external index of several public inboxes, see mirroring instructions on how to clone and mirror all data and code used by this external index.