Help! I lost my C4 states :-)

Help! I lost my C4 states :-)

[Stefan Seyfried]

Hi,

Just curious:
i'm tracking Linus-bk-head quite closely. In (late?) January,
i was able to get my DELL D600 into C4. It worked fine and it
even got me an estimated 15-30 extra minutes out of the battery
(usually it lasts ~3.5 hours, the projected time with C4 was more
like almost 4 hours). Although this is great (it is more than i
ever got from laptop_mode and spinning down the harddisk), it
suddenly stopped working in February. Now the machine almost
always stays in C2, it hardly enters C3 or C4.

seife@strolchi:~> cat /proc/acpi/processor/CPU0/power
active state:            C2
max_cstate:              C8
bus master activity:     10000500
states:
    C1:                  type[C1] promotion[C2] demotion[--] latency[001] usage[00000010]
   *C2:                  type[C2] promotion[C3] demotion[C1] latency[001] usage[107659165]
    C3:                  type[C3] promotion[C4] demotion[C2] latency[085] usage[03505424]
    C4:                  type[C3] promotion[--] demotion[C3] latency[185] usage[06231087]

There is always some bus master activity which i have not been able to track
down. I tried unloading almost all modules, but it did not help. Of course
there is no sound running, no USB mouse attached etc. I have quiesced the
harddisk with laptop_mode and mounting filesystems noatime and commit=300.

The only thing that i can think of what has changed in my setup is that we
changed to modular IDE device drivers, but i cannot really imagine how that
would produce more bus master activity.

Any hints on how to track down what is causing bus master activity all the
time?
Has the calculation algorithm changed? Any default settings i could tweak?
-- 
Stefan Seyfried



-------------------------------------------------------
SF email is sponsored by - The IT Product Guide
Read honest & candid reviews on hundreds of IT Products from real users.
Discover which products truly live up to the hype. Start reading now.
http://ads.osdn.com/?ad_id=6595&alloc_id=14396&op=click

Re: Help! I lost my C4 states :-)

[Len Brown]

On Mon, 2005-02-21 at 03:12, Stefan Seyfried wrote:
> Hi,
> 
> Just curious:
> i'm tracking Linus-bk-head quite closely. In (late?) January,
> i was able to get my DELL D600 into C4. It worked fine and it
> even got me an estimated 15-30 extra minutes out of the battery
> (usually it lasts ~3.5 hours, the projected time with C4 was more
> like almost 4 hours). Although this is great (it is more than i
> ever got from laptop_mode and spinning down the harddisk), it
> suddenly stopped working in February. Now the machine almost
> always stays in C2, it hardly enters C3 or C4.
> 
> seife@strolchi:~> cat /proc/acpi/processor/CPU0/power
> active state:            C2
> max_cstate:              C8
> bus master activity:     10000500
> states:
>     C1:                  type[C1] promotion[C2] demotion[--] latency[001] usage[00000010]
>    *C2:                  type[C2] promotion[C3] demotion[C1] latency[001] usage[107659165]
>     C3:                  type[C3] promotion[C4] demotion[C2] latency[085] usage[03505424]
>     C4:                  type[C3] promotion[--] demotion[C3] latency[185] usage[06231087]
> 
> There is always some bus master activity which i have not been able to track
> down. I tried unloading almost all modules, but it did not help. Of course
> there is no sound running, no USB mouse attached etc. I have quiesced the
> harddisk with laptop_mode and mounting filesystems noatime and commit=300.
> 
> The only thing that i can think of what has changed in my setup is that we
> changed to modular IDE device drivers, but i cannot really imagine how that
> would produce more bus master activity.
> 
> Any hints on how to track down what is causing bus master activity all the
> time?
> Has the calculation algorithm changed? Any default settings i could tweak?

Did you notice a battery life decrease with this decrease in C4
residency?

On January 20th we pushed Dominik's jiffies based bm_activity patch
in drivers/acpi/processor_idle.c  You might find that backing out
that file a couple of revisions restores the previous behaviour.

Just after that, I added a tunable that you can tweak --
with 1000 HZ, its default is all bits enabled, but
you can reduce this bit mask to make the driver
ignore less recent history:

# echo 0 > /sys/module/processor/parameters/bm_history

Will cause the processor driver to ignore the bus master activity
completely.  Note that this could cause a performance hit if
there is BM activity since will just pop the system out of C4
and no instructions are executed during the transition out of C4.

cheers,
-Len

ps.
If you're tracking battery life, I'd be interested to know if
you notice a difference between when you build with HZ=100 and HZ=1000.




-------------------------------------------------------
SF email is sponsored by - The IT Product Guide
Read honest & candid reviews on hundreds of IT Products from real users.
Discover which products truly live up to the hype. Start reading now.
http://ads.osdn.com/?ad_id=6595&alloc_id=14396&op=click

Re: Help! I lost my C4 states :-)

[Pavel Machek]

Hi!

> ps.
> If you're tracking battery life, I'd be interested to know if
> you notice a difference between when you build with HZ=100 and HZ=1000.

At one point I did some measurements using internal battery
ampermeter... lets see:

Arima (youknowwhat)64 processor:
idle machine at 800MHz, max backlight: 22.2 W 
                        min backlight: 17 W   
sleeping harddrive:                   -0.6 W
HZ=100:                               -0.8 W

So yes, HZ=100 and HZ=1000 difference is quite important.
								Pavel
-- 
People were complaining that M$ turns users into beta-testers...
...jr ghea gurz vagb qrirybcref, naq gurl frrz gb yvxr vg gung jnl!


-------------------------------------------------------
SF email is sponsored by - The IT Product Guide
Read honest & candid reviews on hundreds of IT Products from real users.
Discover which products truly live up to the hype. Start reading now.
http://ads.osdn.com/?ad_id=6595&alloc_id=14396&op=click

Re: Help! I lost my C4 states :-)

[Stefan Seyfried]

On Thu, Mar 03, 2005 at 04:38:49PM -0500, Len Brown wrote:
> On Mon, 2005-02-21 at 03:12, Stefan Seyfried wrote:

> > Any hints on how to track down what is causing bus master activity all the
> > time?
> > Has the calculation algorithm changed? Any default settings i could tweak?
> 
> Did you notice a battery life decrease with this decrease in C4
> residency?

Yes, although i lately run the battery down only on rare occasions, but the
"projected" runtime with mostly C4 active was almost 4 hours while now it is
a bit more than 3h30. I do no longer spin down the disk, since my dell locks
up hard sometimes when the disk should spin up again, but the C4 was at least
as big a gain than doing really aggressive disk spindown, (including 
laptop_mode etc, my disk was _really_ spun down for 10 minutes between short
spinups).

In fact, i noticed that C4 started working since i noticed that my battery
started to last longer. When i still spun down the disk, this was hardly
noticeable.
 
> On January 20th we pushed Dominik's jiffies based bm_activity patch
> in drivers/acpi/processor_idle.c  You might find that backing out
> that file a couple of revisions restores the previous behaviour.
> 
> Just after that, I added a tunable that you can tweak --
> with 1000 HZ, its default is all bits enabled, but
> you can reduce this bit mask to make the driver
> ignore less recent history:
> 
> # echo 0 > /sys/module/processor/parameters/bm_history
> 
> Will cause the processor driver to ignore the bus master activity
> completely.  Note that this could cause a performance hit if
> there is BM activity since will just pop the system out of C4
> and no instructions are executed during the transition out of C4.

I tried fiddling with this parameter, but will try again. I think i never
set it to zero, only 0xffff or something like that.
> 
> cheers,
> -Len
> 
> ps.
> If you're tracking battery life, I'd be interested to know if
> you notice a difference between when you build with HZ=100 and HZ=1000.

I'll do it as soon as Andrea Arcangely puts back the "select HZ Patch" back
into the SUSE kernel. Unfortunately i mostly have to run stock compiled
kernels just to be able to file "valid" bugreports :-)
-- 
Stefan Seyfried


-------------------------------------------------------
SF email is sponsored by - The IT Product Guide
Read honest & candid reviews on hundreds of IT Products from real users.
Discover which products truly live up to the hype. Start reading now.
http://ads.osdn.com/?ad_id=6595&alloc_id=14396&op=click