does cpu_idle() halt CPU?

does cpu_idle() halt CPU?

[Jun Sun]

I am new to acpi, but I figure this question is probably
beyond acpi-support list.

I instrumented acpi_processor_idle() routine to see
what states the CPU is in.  It turns out it spent most
of the time in C2 state and some times in C1, but never
in C3.

However, acpi_processor_power_activate() routine  does not
do anything if the new state is _not_ C3.  So is that true
CPU still keep spinning all the time?

I did a rough comparison.  It appears my battery life
under Linux is about 75%  of that under Windows, even after
I installed acpi, did a bunch of other stuff (swapoff,
EXT3 -> EXT2, etc).

Jun


-------------------------------------------------------
This SF.net email is sponsored by:
The Definitive IT and Networking Event. Be There!
NetWorld+Interop Las Vegas 2003 -- Register today!
http://ads.sourceforge.net/cgi-bin/redirect.pl?keyn0001en

RE: does cpu_idle() halt CPU?

[Grover, Andrew]

> From: Jun Sun [mailto:jsun-vWJf10E4RRusTnJN9+BGXg@public.gmane.org] 
> I am new to acpi, but I figure this question is probably
> beyond acpi-support list.
> 
> I instrumented acpi_processor_idle() routine to see
> what states the CPU is in.  It turns out it spent most
> of the time in C2 state and some times in C1, but never
> in C3.
> 
> However, acpi_processor_power_activate() routine  does not
> do anything if the new state is _not_ C3.  So is that true
> CPU still keep spinning all the time?
> 
> I did a rough comparison.  It appears my battery life
> under Linux is about 75%  of that under Windows, even after
> I installed acpi, did a bunch of other stuff (swapoff,
> EXT3 -> EXT2, etc).

What power states does your CPU support?
- What does a cat of /proc/acpi/processor/*/power say?
- What does the ACPI: Processor line say in your dmesg say?

What kind of laptop do you have? Does it support SpeedStep, PowerNow, or
the like?
Do you have any USB devices attached?

Regards -- Andy


-------------------------------------------------------
This SF.net email is sponsored by:
The Definitive IT and Networking Event. Be There!
NetWorld+Interop Las Vegas 2003 -- Register today!
http://ads.sourceforge.net/cgi-bin/redirect.pl?keyn0001en

Re: does cpu_idle() halt CPU?

[Jun Sun]

On Fri, Mar 28, 2003 at 02:28:41PM -0800, Grover, Andrew wrote:
> 
> What power states does your CPU support?
> - What does a cat of /proc/acpi/processor/*/power say?

[root@localhost root]# cat /proc/acpi/processor/CPU0/power
active state:            C2
default state:           C1
bus master activity:     ffffffff
states:
    C1:                  promotion[C2] demotion[--] latency[000] usage[00000690]   *C2:                  promotion[C3] demotion[C1] latency[010] usage[00196669]    C3:                  promotion[--] demotion[C2] latency[130] usage[00000000]

> - What does the ACPI: Processor line say in your dmesg say?
>
ACPI: RSDP (v000 FUJ                        ) @ 0x000f6200
ACPI: RSDT (v001 FUJ    CLOVE    00263.00000) @ 0x1f6ea3fd
ACPI: FADT (v001 FUJ    CLOVE    00263.00000) @ 0x1f6efb8c
ACPI: DSDT (v001 FUJ    CLOVE    00263.00000) @ 0x00000000
ACPI: BIOS passes blacklist
...
Intel machine check reporting enabled on CPU#0.
CPU:     After generic, caps: 0383f9ff 00000000 00000000 00000000
CPU:             Common caps: 0383f9ff 00000000 00000000 00000000
CPU: Intel Mobile Intel(R) Pentium(R) III CPU - M  1200MHz stepping 04
Enabling fast FPU save and restore... done.
...
 
> What kind of laptop do you have? Does it support SpeedStep, PowerNow, or
> the like?

My laptop is Fujitsu S6110.  CPU supports speedstep.

> Do you have any USB devices attached?
>

Not in this particular run.

Jun 


-------------------------------------------------------
This SF.net email is sponsored by:
The Definitive IT and Networking Event. Be There!
NetWorld+Interop Las Vegas 2003 -- Register today!
http://ads.sourceforge.net/cgi-bin/redirect.pl?keyn0001en

RE: does cpu_idle() halt CPU?

[Grover, Andrew]

> From: Jun Sun [mailto:jsun-vWJf10E4RRusTnJN9+BGXg@public.gmane.org] 

> [root@localhost root]# cat /proc/acpi/processor/CPU0/power
> active state:            C2
> default state:           C1
> bus master activity:     ffffffff
> states:
>     C1:                  promotion[C2] demotion[--] 
> latency[000] usage[00000690]   *C2:                  
> promotion[C3] demotion[C1] latency[010] usage[00196669]    
> C3:                  promotion[--] demotion[C2] latency[130] 
> usage[00000000]

Your system either doesn't support C3, or the constant bus master
activity is keeping it out of C3. USB can cause this, but also AC97.

> > - What does the ACPI: Processor line say in your dmesg say?

There should be a line that says, "ACPI: Processor found (x perf states
y throttle)" Is that there? It may be occurring later on in the boot
process.

> > What kind of laptop do you have? Does it support SpeedStep, 
> PowerNow, or
> > the like?
> 
> My laptop is Fujitsu S6110.  CPU supports speedstep.

Getting into the lowest SpeedStep state will help battery life. You may
want to investigate the cpufreq project.

Regards -- Andy


-------------------------------------------------------
This SF.net email is sponsored by:
The Definitive IT and Networking Event. Be There!
NetWorld+Interop Las Vegas 2003 -- Register today!
http://ads.sourceforge.net/cgi-bin/redirect.pl?keyn0001en

Re: does cpu_idle() halt CPU?

[Jun Sun]

On Fri, Mar 28, 2003 at 02:43:32PM -0800, Grover, Andrew wrote:
> 
> Your system either doesn't support C3, or the constant bus master
> activity is keeping it out of C3. USB can cause this, but also AC97.
>

No USB and AC97.  Have 802.11 card working.  Not sure if this matters.

 
> > > - What does the ACPI: Processor line say in your dmesg say?
> 
> There should be a line that says, "ACPI: Processor found (x perf states
> y throttle)" Is that there? It may be occurring later on in the boot
> process.
>

I found this line:
 
ACPI: Processor [CPU0] (supports C1 C2 C3, 2 performance states)

> > > What kind of laptop do you have? Does it support SpeedStep, 
> > PowerNow, or
> > > the like?
> > 
> > My laptop is Fujitsu S6110.  CPU supports speedstep.
> 
> Getting into the lowest SpeedStep state will help battery life. You may
> want to investigate the cpufreq project.
> 

So my previous conclusion about C1/C2 not saving power is right?

Thanks a lot.

Jun


-------------------------------------------------------
This SF.net email is sponsored by:
The Definitive IT and Networking Event. Be There!
NetWorld+Interop Las Vegas 2003 -- Register today!
http://ads.sourceforge.net/cgi-bin/redirect.pl?keyn0001en

RE: does cpu_idle() halt CPU?

[Stephan Krings]

Hello,

> > From: Jun Sun [mailto:jsun-vWJf10E4RRusTnJN9+BGXg@public.gmane.org] 
> > I am new to acpi, but I figure this question is probably
> > beyond acpi-support list.
> > 
> > I instrumented acpi_processor_idle() routine to see
> > what states the CPU is in.  It turns out it spent most
> > of the time in C2 state and some times in C1, but never
> > in C3.

I recently discovered that on my notebook the CPU state (and therefore
power consumption) is highly dependent on the drivers loaded. There
are at least 2 drivers, which practically seem to inhibit the CPU from
entering C3: The usb-ohci module (must only be loaded, no USB devices
need to be attached) and DRI (direct rendering for XFree86).

When these are enabled the CPU never enters C3. If I disable them, the
CPU is in C3 most of the time, which can easily be monitored by
something like:

watch --interval=1 cat /proc/acpi/processor/C000/power

Apart from ACPI itself, there is also quite a lot of work required to
make individual drivers more power management aware.

Depending on your hardware and setup you might consider disabling some
drivers for hardware you don't actually need. This might greatly
increase your battery life.

Regards,

Stephan


-------------------------------------------------------
This SF.net email is sponsored by:
The Definitive IT and Networking Event. Be There!
NetWorld+Interop Las Vegas 2003 -- Register today!
http://ads.sourceforge.net/cgi-bin/redirect.pl?keyn0001en

Re: does cpu_idle() halt CPU?

[Ducrot Bruno]

On Fri, Mar 28, 2003 at 03:02:18PM -0800, Jun Sun wrote:
> On Fri, Mar 28, 2003 at 02:43:32PM -0800, Grover, Andrew wrote:
> > 
> > Your system either doesn't support C3, or the constant bus master
> > activity is keeping it out of C3. USB can cause this, but also AC97.
> >
> 
> No USB and AC97.  Have 802.11 card working.  Not sure if this matters.

Who knows..

> 
>  
> > > > - What does the ACPI: Processor line say in your dmesg say?
> > 
> > There should be a line that says, "ACPI: Processor found (x perf states
> > y throttle)" Is that there? It may be occurring later on in the boot
> > process.
> >
> 
> I found this line:
>  
> ACPI: Processor [CPU0] (supports C1 C2 C3, 2 performance states)
> 
> > > > What kind of laptop do you have? Does it support SpeedStep, 
> > > PowerNow, or
> > > > the like?
> > > 
> > > My laptop is Fujitsu S6110.  CPU supports speedstep.
> > 
> > Getting into the lowest SpeedStep state will help battery life. You may
> > want to investigate the cpufreq project.
> > 
> 
> So my previous conclusion about C1/C2 not saving power is right?
> 

No.  C2 will give you some power saving (try to disable explicitely and
you will see a decrease of the battery life).

You forgot at least disk activities especially if ext3, which
prevent almost all IDE drivers to go to idle state.
But that will not explain why you can not go to C3 though.

Cheers,

-- 
Ducrot Bruno

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


-------------------------------------------------------
This SF.net email is sponsored by:
The Definitive IT and Networking Event. Be There!
NetWorld+Interop Las Vegas 2003 -- Register today!
http://ads.sourceforge.net/cgi-bin/redirect.pl?keyn0001en

Re: does cpu_idle() halt CPU?

[Jun Sun]

On Sat, Mar 29, 2003 at 11:10:18AM +0100, Stephan Krings wrote:
> 
> Hello,
> 
> > > From: Jun Sun [mailto:jsun-vWJf10E4RRusTnJN9+BGXg@public.gmane.org] 
> > > I am new to acpi, but I figure this question is probably
> > > beyond acpi-support list.
> > > 
> > > I instrumented acpi_processor_idle() routine to see
> > > what states the CPU is in.  It turns out it spent most
> > > of the time in C2 state and some times in C1, but never
> > > in C3.
> 
> I recently discovered that on my notebook the CPU state (and therefore
> power consumption) is highly dependent on the drivers loaded. There
> are at least 2 drivers, which practically seem to inhibit the CPU from
> entering C3: The usb-ohci module (must only be loaded, no USB devices
> need to be attached) and DRI (direct rendering for XFree86).
>

This turns out to be key.  BTW, I did an experiement, usb-uhci has
the same effect too.

Thanks.

Jun

P.S., Converting EXT3 back to EXT2 is the first thing I did.  I was 
puzzled by the fact that disk never spins down and traced it back to
journald.



-------------------------------------------------------
This SF.net email is sponsored by:
The Definitive IT and Networking Event. Be There!
NetWorld+Interop Las Vegas 2003 -- Register today!
http://ads.sourceforge.net/cgi-bin/redirect.pl?keyn0001en

RE: does cpu_idle() halt CPU?

[Grover, Andrew]

> From: Jun Sun [mailto:jsun-vWJf10E4RRusTnJN9+BGXg@public.gmane.org] 
> > I recently discovered that on my notebook the CPU state 
> (and therefore
> > power consumption) is highly dependent on the drivers loaded. There
> > are at least 2 drivers, which practically seem to inhibit 
> the CPU from
> > entering C3: The usb-ohci module (must only be loaded, no 
> USB devices
> > need to be attached) and DRI (direct rendering for XFree86).
> >
> 
> This turns out to be key.  BTW, I did an experiement, usb-uhci has
> the same effect too.

This is a bug.

I fixed the other uhci driver but didn't fix usb-uhci, a couple years
ago. Basically, the HC should go into Global Suspend when no USB devices
are connected to it, and wake up when one is connected.

Using the other uhci driver as a model, this might be a good task for
someone interested in hacking the kernel to get their feet wet. ;-)

Regards -- Andy


-------------------------------------------------------
This SF.net email is sponsored by: ValueWeb: 
Dedicated Hosting for just $79/mo with 500 GB of bandwidth! 
No other company gives more support or power for your dedicated server
http://click.atdmt.com/AFF/go/sdnxxaff00300020aff/direct/01/

Re: does cpu_idle() halt CPU?

[Jan Rychter]

>>>>> "DB" == Ducrot Bruno <ducrot-kk6yZipjEM5g9hUCZPvPmw@public.gmane.org>:
[...]
 DB> You forgot at least disk activities especially if ext3, which
 DB> prevent almost all IDE drivers to go to idle state.  But that will
 DB> not explain why you can not go to C3 though.

By the way, let me ask here: does anybody know what would be the best
filesystem to use on a laptop, power saving being the #1 priority? Are
all journaling filesystems this bad?

There was a project to teach the Linux kernel about drive spindown and
manage drive accesses accordingly -- but last time I saw their docs they
said that all journaling filesystems pose serious problems...

USB preventing the machine from going into C3 is a serious problem. I
still don't understand if that's a hardware thing or can we have a more
reasonable HCI implementation that doesn't use DMA that much. How does
Windows deal with it?

--J.


-------------------------------------------------------
This SF.net email is sponsored by: ValueWeb: 
Dedicated Hosting for just $79/mo with 500 GB of bandwidth! 
No other company gives more support or power for your dedicated server
http://click.atdmt.com/AFF/go/sdnxxaff00300020aff/direct/01/

Re: Re: does cpu_idle() halt CPU?

[Matthew Keller]

SGI's XFS is wonderful, AFAICT. Just make sure you have the latest tools
and whatnot.

On Wed, 2003-04-02 at 02:05, Jan Rychter wrote:
> >>>>> "DB" == Ducrot Bruno <ducrot-kk6yZipjEM5g9hUCZPvPmw@public.gmane.org>:
> [...]
>  DB> You forgot at least disk activities especially if ext3, which
>  DB> prevent almost all IDE drivers to go to idle state.  But that will
>  DB> not explain why you can not go to C3 though.
> 
> By the way, let me ask here: does anybody know what would be the best
> filesystem to use on a laptop, power saving being the #1 priority? Are
> all journaling filesystems this bad?
> 
> There was a project to teach the Linux kernel about drive spindown and
> manage drive accesses accordingly -- but last time I saw their docs they
> said that all journaling filesystems pose serious problems...
> 
> USB preventing the machine from going into C3 is a serious problem. I
> still don't understand if that's a hardware thing or can we have a more
> reasonable HCI implementation that doesn't use DMA that much. How does
> Windows deal with it?
> 
> --J.
> 
> 
> -------------------------------------------------------
> This SF.net email is sponsored by: ValueWeb: 
> Dedicated Hosting for just $79/mo with 500 GB of bandwidth! 
> No other company gives more support or power for your dedicated server
> http://click.atdmt.com/AFF/go/sdnxxaff00300020aff/direct/01/
> _______________________________________________
> Acpi-devel mailing list
> Acpi-devel-5NWGOfrQmneRv+LV9MX5uipxlwaOVQ5f@public.gmane.org
> https://lists.sourceforge.net/lists/listinfo/acpi-devel
-- 
Matthew Keller
Enterprise Systems Analyst
Computing & Technology Services
State University of New York @ Potsdam
Potsdam, NY USA
http://mattwork.potsdam.edu/



-------------------------------------------------------
This SF.net email is sponsored by: ValueWeb: 
Dedicated Hosting for just $79/mo with 500 GB of bandwidth! 
No other company gives more support or power for your dedicated server
http://click.atdmt.com/AFF/go/sdnxxaff00300020aff/direct/01/

Re: Re: does cpu_idle() halt CPU?

[Charl P. Botha]

On Wed, Apr 02, 2003 at 09:05:06AM +0200, Jan Rychter wrote:
> >>>>> "DB" == Ducrot Bruno <ducrot-kk6yZipjEM5g9hUCZPvPmw@public.gmane.org>:
> [...]
>  DB> You forgot at least disk activities especially if ext3, which
>  DB> prevent almost all IDE drivers to go to idle state.  But that will
>  DB> not explain why you can not go to C3 though.
> 
> By the way, let me ask here: does anybody know what would be the best
> filesystem to use on a laptop, power saving being the #1 priority? Are
> all journaling filesystems this bad?

I'm having great results with ext3 by making sure that all my filesystems
are mounted with the noatime option.

-- 
charl p. botha http://cpbotha.net/ http://visualisation.tudelft.nl/


-------------------------------------------------------
This SF.net email is sponsored by: ValueWeb: 
Dedicated Hosting for just $79/mo with 500 GB of bandwidth! 
No other company gives more support or power for your dedicated server
http://click.atdmt.com/AFF/go/sdnxxaff00300020aff/direct/01/

Re: Re: does cpu_idle() halt CPU?

[Karol Kozimor]

Thus wrote Matthew Keller:
> SGI's XFS is wonderful, AFAICT. Just make sure you have the latest tools
> and whatnot.

Are you sure? I am using XFS on both my partitions, both mounted with
noatime, I told syslog not to sync upon write, yet the disk still keeps
spinning up arbitrarily (usually after 30 - 90 seconds). Or perhaps the
cause lies somewhere else?
Best regards,

-- 
Karol 'sziwan' Kozimor
sziwan-DETuoxkZsSqrDJvtcaxF/A@public.gmane.org


-------------------------------------------------------
This SF.net email is sponsored by: ValueWeb: 
Dedicated Hosting for just $79/mo with 500 GB of bandwidth! 
No other company gives more support or power for your dedicated server
http://click.atdmt.com/AFF/go/sdnxxaff00300020aff/direct/01/

Re: Re: does cpu_idle() halt CPU?

[Matthew Keller]

On Thu, 2003-04-03 at 01:47, Karol Kozimor wrote:
> Thus wrote Matthew Keller:
> > SGI's XFS is wonderful, AFAICT. Just make sure you have the latest tools
> > and whatnot.
> 
> Are you sure? I am using XFS on both my partitions, both mounted with
> noatime, I told syslog not to sync upon write, yet the disk still keeps
> spinning up arbitrarily (usually after 30 - 90 seconds). Or perhaps the
> cause lies somewhere else?
> Best regards,

I've got 4 partitions (root, /usr, /home and /var/log) and everything is
copacetic. One thing to note is that I, uh, don't use syslog on my
laptop as disabling it saves my battery life by 10-15%. Try disabling it
and seeing where that leads, perhaps?

-- 
Matthew Keller
Enterprise Systems Analyst
Computing & Technology Services
State University of New York @ Potsdam
Potsdam, NY USA
http://mattwork.potsdam.edu/



-------------------------------------------------------
This SF.net email is sponsored by: ValueWeb: 
Dedicated Hosting for just $79/mo with 500 GB of bandwidth! 
No other company gives more support or power for your dedicated server
http://click.atdmt.com/AFF/go/sdnxxaff00300020aff/direct/01/

Re: does cpu_idle() halt CPU?

[Jan Rychter]

>>>>> "Matthew" == Matthew Keller <kellermg-aGob3lNJgTKVc3sceRu5cw@public.gmane.org>:
 Matthew> On Thu, 2003-04-03 at 01:47, Karol Kozimor wrote:
 >> Thus wrote Matthew Keller:
 > SGI's XFS is wonderful, AFAICT. Just make sure you have the latest
 > tools and whatnot.
 >>
 >> Are you sure? I am using XFS on both my partitions, both mounted
 >> with noatime, I told syslog not to sync upon write, yet the disk
 >> still keeps spinning up arbitrarily (usually after 30 - 90
 >> seconds). Or perhaps the cause lies somewhere else?  Best regards,

 Matthew> I've got 4 partitions (root, /usr, /home and /var/log) and
 Matthew> everything is copacetic. One thing to note is that I, uh,
 Matthew> don't use syslog on my laptop as disabling it saves my battery
 Matthew> life by 10-15%. Try disabling it and seeing where that leads,
 Matthew> perhaps?

noflushd (http://noflushd.sourceforge.net/ is what I was talking about
when I mentioned a project to hold disk writes when the disk is spun
down.

Unfortunately, the project info page says it doesn't help with
reiserfs. It also seems to be stagnant (but perhaps it just works so
well?).

--J.


-------------------------------------------------------
This SF.net email is sponsored by: ValueWeb: 
Dedicated Hosting for just $79/mo with 500 GB of bandwidth! 
No other company gives more support or power for your dedicated server
http://click.atdmt.com/AFF/go/sdnxxaff00300020aff/direct/01/

Re: does cpu_idle() halt CPU?

[Ben Slusky]

On Mon, 31 Mar 2003 10:15:01 -0800, Grover, Andrew wrote:
> > From: Jun Sun [mailto:jsun-vWJf10E4RRusTnJN9+BGXg@public.gmane.org] 
> > This turns out to be key.  BTW, I did an experiement, usb-uhci has
> > the same effect too.
> 
> This is a bug.
> 
> I fixed the other uhci driver but didn't fix usb-uhci, a couple years
> ago. Basically, the HC should go into Global Suspend when no USB devices
> are connected to it, and wake up when one is connected.

The usb-ohci driver seems to have the same problem; my Libretto L1 won't
go into C3 when the module is loaded. I took a look at the driver but
of course it's very different from either of the uhci drivers, and way
over the head of a humble sysadmin such as myself... :-(

-- 
Ben Slusky                      | "Don't let it end like this.
sluskyb-8e/Pmov9ym4dnm+yROfE0A@public.gmane.org              |  Tell them I said something."
sluskyb-k/7z6JowIC0cuXOPmUuyGA@public.gmane.org          |               -last words of
PGP keyID ADA44B3B              |                Pancho Villa


-------------------------------------------------------
This SF.net email is sponsored by: Etnus, makers of TotalView, The debugger 
for complex code. Debugging C/C++ programs can leave you feeling lost and 
disoriented. TotalView can help you find your way. Available on major UNIX 
and Linux platforms. Try it free. www.etnus.com