| View previous topic :: View next topic |
| Author |
Message |
manuels
Advocate
Joined: 22 Nov 2003
Posts: 2146
Location: Europe
|
 Posted: Tue Oct 14, 2008 8:53 pm Post subject: Powertop: reduce count of wakeups |
 |
|
Hi,
I'm getting this output of powertop:
| Code: | Top causes for wakeups:
43.8% (226.6) <interrupt> : extra timer interrupt
25.0% (129.2) <interrupt> : nvidia, wifi0
19.5% (100.6) <interrupt> : eth0
5.1% ( 26.4) firefox : futex_wait (hrtimer_wakeup)
2.1% ( 10.8) <interrupt> : ide0
0.8% ( 4.2) <interrupt> : sata_nv
0.7% ( 3.8) kded : schedule_timeout (process_timeout)
0.4% ( 2.0) <kernel IPI> : Rescheduling interrupts
0.3% ( 1.6) wpa_supplicant : schedule_timeout (process_timeout)
0.3% ( 1.4) X : do_setitimer (it_real_fn)
0.2% ( 1.0) X : nv_start_rc_timer (nv_kern_rc_timer)
0.2% ( 1.0) <kernel core> : ath_detach (ath_intr)
0.2% ( 1.0) dhcdbd : schedule_timeout (process_timeout)
0.2% ( 0.8) yakuake : schedule_timeout (process_timeout)
0.2% ( 0.8) <kernel core> : neigh_table_init_no_netlink (neigh_periodic_timer)
|
I'm using a tickless system but there are still high number of these 'extra timer interrupt'.
Does anybody of you have a hint for me to minimize this number of wakeups?
Some information about my system:
| Code: | | Linux roadrunner 2.6.26-tuxonice #5 SMP PREEMPT Sun Sep 28 22:29:06 CEST 2008 x86_64 AMD Turion(tm) 64 X2 Mobile Technology TL-50 AuthenticAMD GNU/Linux |
my kernel config
_________________
Build your own live cd with catalyst 2.0! |
|
| Back to top |
|
 |
dmpogo
Advocate
Joined: 02 Sep 2004
Posts: 3468
Location: Canada
|
| Posted: Wed Oct 15, 2008 1:45 am Post subject: Re: Powertop: reduce count of wakeups |
|
|
No hints, I recently googled extensively for this question and found nothing, besides that they are still present.
They are, however, < 10% of what I had in timer interrupts before going tickless |
|
| Back to top |
|
|
Hypnos
Advocate
Joined: 18 Jul 2002
Posts: 2889
Location: Omnipresent
|
| Posted: Wed Mar 11, 2009 9:05 am Post subject: |
|
|
Are you using hpet? If not, enable it in your kernel.
'dmesg | grep -i hpet' should give a bunch of stuff. My output:
| Code: | ACPI: HPET 7E6CBD16, 0038 (r1 LENOVO TP-7L 2210 LNVO 1)
ACPI: HPET id: 0x8086a201 base: 0xfed00000
hpet clockevent registered
TSC calibrated against HPET
hpet0: at MMIO 0xfed00000, IRQs 2, 8, 0
hpet0: 3 64-bit timers, 14318180 Hz
Time: hpet clocksource has been installed.
hpet_resources: 0xfed00000 is busy |
_________________
Personal overlay | Simple backup scheme |
|
| Back to top |
|
|
manuels
Advocate
Joined: 22 Nov 2003
Posts: 2146
Location: Europe
|
| Posted: Mon Apr 27, 2009 8:40 pm Post subject: |
|
|
Sorry for my late response.
Yes, I am using it:
| Code: | ACPI: HPET 3FF69F42, 0038 (r1 PTLTD HPETTBL 6040000 LTP 1)
ACPI: HPET id: 0x10de8201 base: 0xfed00000
hpet clockevent registered
HPET: 3 timers in total, 0 timers will be used for per-cpu timer
hpet0: at MMIO 0xfed00000, IRQs 2, 8, 31
hpet0: 3 comparators, 32-bit 25.000000 MHz counter
CE: hpet increasing min_delta_ns to 15000 nsec
CE: hpet increasing min_delta_ns to 22500 nsec
|
_________________
Build your own live cd with catalyst 2.0! |
|
| Back to top |
|
|
Hypnos
Advocate
Joined: 18 Jul 2002
Posts: 2889
Location: Omnipresent
|
| Posted: Tue Apr 28, 2009 1:29 am Post subject: |
|
|
My output on battery power:
| Code: | Cn Avg residency P-states (frequencies)
C0 (cpu running) ( 5.9%) 2.21 Ghz 3.8%
C1 0.0ms ( 0.0%) 2.21 Ghz 0.0%
C2 0.1ms ( 0.0%) 1.60 Ghz 0.1%
C3 3.9ms (94.1%) 800 Mhz 96.1%
Wakeups-from-idle per second : 244.5 interval: 10.0s
Power usage (ACPI estimate): 15.1W (3.7 hours)
Top causes for wakeups:
22.3% ( 50.0) hdapsd : hdaps_mousedev_open (hdaps_mousedev_poll)
20.9% ( 46.9) <interrupt> : uhci_hcd:usb4, HDA Intel, ohci1394
17.6% ( 39.5) <kernel IPI> : Rescheduling interrupts
11.5% ( 25.9) epiphany : futex_wait (hrtimer_wakeup)
5.7% ( 12.7) <interrupt> : iwl3945
5.1% ( 11.5) skype : schedule_hrtimeout_range (hrtimer_wakeup)
|
I have implemented all the hints that powertop gives me, and I've had significant improvement in "rescheduling interrupts" by upgrading to kernel 2.6.29. If I disable HDAPS I attain the target of 95% residency in the CPU state C3. (Interestingly, I attain the goal easily while on AC -- I wonder if some ACPI thing is causing interrupts while on battery power.)
If you're not running a recent kernel, you might give it a try ...
Also, I don't understand your dmesg output entirely -- HPET is detected, but I wonder if it's actually being used ...
_________________
Personal overlay | Simple backup scheme |
|
| Back to top |
|
|
theotherjoe
Guru
Joined: 22 Nov 2003
Posts: 393
|
| Posted: Tue Apr 28, 2009 6:40 am Post subject: |
|
|
Hypnos,
here is how to find out about your current clocksource
# cat /sys/devices/system/clocksource/clocksource0/current_clocksource
and via
# cat /sys/devices/system/clocksource/clocksource0/available_clocksource
have a look at the different possible clocks to employ |
|
| Back to top |
|
|
manuels
Advocate
Joined: 22 Nov 2003
Posts: 2146
Location: Europe
|
| Posted: Tue Apr 28, 2009 8:58 pm Post subject: |
|
|
Hi,
| Hypnos wrote: | | [...]If you're not running a recent kernel, you might give it a try ... |
I'm using suspend to ram/disk with tuxonice-sources and I just see that there is a new 2.6.29 version.
Ok, I'll compile it tomorrow and see if it will reduce the wakeups.
| Quote: | Also, I don't understand your dmesg output entirely -- HPET is detected, but I wonder if it's actually being used ...
| Quote: | here is how to find out about your current clocksource
# cat /sys/devices/system/clocksource/clocksource0/current_clocksource |
|
That gives me
| Code: | # cat /sys/devices/system/clocksource/clocksource0/current_clocksource
hpet |
So it seems to be enabled.
Nevertheless, I'll give the .29 sources a try. I let you know if there are any news.
Cheers,
Manuel
_________________
Build your own live cd with catalyst 2.0! |
|
| Back to top |
|
|
Hypnos
Advocate
Joined: 18 Jul 2002
Posts: 2889
Location: Omnipresent
|
| Posted: Wed May 20, 2009 11:14 am Post subject: |
|
|
FWIW, I upgraded from acpid-1.0.6-r1 to acpid-1.0.10_p3 and now my C3 residency is 97-98%, even with wireless and HDAPS running.
This upgrade finally moves acpid to the netlink interface from the old /proc/acpi/event -- perhaps this method of communication is more efficient.
_________________
Personal overlay | Simple backup scheme |
|
| Back to top |
|
|
manuels
Advocate
Joined: 22 Nov 2003
Posts: 2146
Location: Europe
|
| Posted: Sun May 24, 2009 1:43 pm Post subject: |
|
|
That didn't work for me:
| Code: | P-states (frequencies)
1.60 Ghz 0.0%
800 Mhz 100.0%
< Detailed C-state information is only available on Mobile CPUs (laptops) >
Wakeups-from-idle per second : 201.1 interval: 10.0s
no ACPI power usage estimate available
Top causes for wakeups:
51.1% (177.0) <interrupt> : wifi0, nvidia
22.1% ( 76.4) <interrupt> : PS/2 keyboard/mouse/touchpad
5.5% ( 19.1) <interrupt> : extra timer interrupt
4.5% ( 15.7) <kernel IPI> : Rescheduling interrupts
3.9% ( 13.6) X : do_setitimer (it_real_fn)
3.7% ( 12.7) firefox : futex_wait (hrtimer_wakeup)
1.9% ( 6.7) kwin : schedule_hrtimeout_range (hrtimer_wakeup)
1.4% ( 5.0) amarok : schedule_hrtimeout_range (hrtimer_wakeup)
0.9% ( 3.2) yakuake : schedule_hrtimeout_range (hrtimer_wakeup)
0.9% ( 3.0) knotify4 : schedule_hrtimeout_range (hrtimer_wakeup)
0.6% ( 2.1) <interrupt> : sata_nv
0.6% ( 2.0) thunderbird-bin : futex_wait (hrtimer_wakeup)
|
Seems that I really must update to a newer kernel - actually, I'm a bit to lazy to do that...
_________________
Build your own live cd with catalyst 2.0! |
|
| Back to top |
|
|
Hypnos
Advocate
Joined: 18 Jul 2002
Posts: 2889
Location: Omnipresent
|
| Posted: Sun May 24, 2009 2:04 pm Post subject: |
|
|
200 wakeups/seconds isn't bad per se -- it depends on what is causing the wakeup. nvidia and wifi are fairly power-hungry 
You really need C-state data to know how these interrupts are hurting the efficiency of your CPU, which is very power hungry. One experiment to try is with and without the wifi enabled.
Also, keyboard/mouse interrupts shouldn't happen unless you are using them ...
_________________
Personal overlay | Simple backup scheme |
|
| Back to top |
|
|
|
Kernel & Hardware
