Re: [Powertop] USB devices get wrong power use values

Re: [Powertop] USB devices get wrong power use values

[Marc MERLIN]

[-- Attachment #1: Type: text/plain, Size: 2288 bytes --]

So, I asked on the linux-usb list, and one thing I quickly found out is that the
Watt value given by powertop 2.4 is way off for USB devices:

Alan Stern wrote the following:

Do you have a bug on powertop 2.4 mis-calculating USB power use, both when 
power is plugged in and when it's not?

Thanks,
Marc

On Sun, Sep 22, 2013 at 12:38:56PM -0400, Alan Stern wrote:
> > > gandalfthegreat:/sys/bus/usb/devices/3-1.6/power# grep . *
> > > active_duration:61227648
> > > async:enabled
> > > autosuspend:2
> > > autosuspend_delay_ms:2000
> > > connected_duration:66830880
> > > control:auto
> > > level:auto
> > > persist:1
> > > runtime_active_kids:0
> > > runtime_active_time:18870052
> > > runtime_enabled:enabled
> > > runtime_status:active
> > > runtime_suspended_time:5324088
> > > runtime_usage:0
> >
> > This all looks correct.
>
> Since then, I've confirmed that I don't have the problem some time after
> reboot. It may be that the device doesn't seem to sleep well after I've used
> it once.
>
> What's interesting, is that I see this when power is plugged in:
>
> Power est.  Events/s    Category       Description
>   8.18 W    100.0%      Device         USB device: Yubico Yubikey II (Yubico)
>   8.13 W    100.0%      Device         USB device: Integrated Camera (Chicony Electronics Co., Ltd.)
>
> Once I go to batteries, I see this:
> Summary: 760.1 wakeups/second,  718.9 GPU ops/seconds, 0.0 VFS ops/sec and 6.9% CPU use
>
> Power est.              Usage       Events/s    Category       Description
>   8.32 W    100.0%                      Device         USB device: Yubico Yubikey II (Yubico)
>   2.52 W     73.3%                      Device         Display backlight
>
> Somehow I know that my Yubikey isn't using 8W, so powertop numbers need to
> be taken with a grain of salt.

I don't know where powertop gets its numbers from.  Perhaps it uses the
value reported by the device (bMaxPower).  That value is only a
maximum; it doesn't change to reflect the actual usage.

-- 
"A mouse is a device used to point at the xterm you want to type in" - A.S.R.
Microsoft is to operating systems ....
                                      .... what McDonalds is to gourmet cooking
Home page: http://marc.merlins.org/

Re: [Powertop] USB devices get wrong power use values

[Sergey Senozhatsky]

[-- Attachment #1: Type: text/plain, Size: 2580 bytes --]

On (09/22/13 16:54), Marc MERLIN wrote:
> So, I asked on the linux-usb list, and one thing I quickly found out is that the
> Watt value given by powertop 2.4 is way off for USB devices:
> 
> Alan Stern wrote the following:
> 
> Do you have a bug on powertop 2.4 mis-calculating USB power use, both when 
> power is plugged in and when it's not?

powertop uses

	/sys/bus/usb/devices/%s/power/active_duration
and
	/sys/bus/usb/devices/%s/power/connected_duration

	-ss

> Thanks,
> Marc
> 
> On Sun, Sep 22, 2013 at 12:38:56PM -0400, Alan Stern wrote:
> > > > gandalfthegreat:/sys/bus/usb/devices/3-1.6/power# grep . *
> > > > active_duration:61227648
> > > > async:enabled
> > > > autosuspend:2
> > > > autosuspend_delay_ms:2000
> > > > connected_duration:66830880
> > > > control:auto
> > > > level:auto
> > > > persist:1
> > > > runtime_active_kids:0
> > > > runtime_active_time:18870052
> > > > runtime_enabled:enabled
> > > > runtime_status:active
> > > > runtime_suspended_time:5324088
> > > > runtime_usage:0
> > >
> > > This all looks correct.
> >
> > Since then, I've confirmed that I don't have the problem some time after
> > reboot. It may be that the device doesn't seem to sleep well after I've used
> > it once.
> >
> > What's interesting, is that I see this when power is plugged in:
> >
> > Power est.  Events/s    Category       Description
> >   8.18 W    100.0%      Device         USB device: Yubico Yubikey II (Yubico)
> >   8.13 W    100.0%      Device         USB device: Integrated Camera (Chicony Electronics Co., Ltd.)
> >
> > Once I go to batteries, I see this:
> > Summary: 760.1 wakeups/second,  718.9 GPU ops/seconds, 0.0 VFS ops/sec and 6.9% CPU use
> >
> > Power est.              Usage       Events/s    Category       Description
> >   8.32 W    100.0%                      Device         USB device: Yubico Yubikey II (Yubico)
> >   2.52 W     73.3%                      Device         Display backlight
> >
> > Somehow I know that my Yubikey isn't using 8W, so powertop numbers need to
> > be taken with a grain of salt.
> 
> I don't know where powertop gets its numbers from.  Perhaps it uses the
> value reported by the device (bMaxPower).  That value is only a
> maximum; it doesn't change to reflect the actual usage.
> 
> -- 
> "A mouse is a device used to point at the xterm you want to type in" - A.S.R.
> Microsoft is to operating systems ....
>                                       .... what McDonalds is to gourmet cooking
> Home page: http://marc.merlins.org/
> 

Re: [Powertop] USB devices get wrong power use values

[Arjan van de Ven]

[-- Attachment #1: Type: text/plain, Size: 679 bytes --]

On 9/22/2013 4:54 PM, Marc MERLIN wrote:
> So, I asked on the linux-usb list, and one thing I quickly found out is that the
> Watt value given by powertop 2.4 is way off for USB devices:

what makes you think the value is off?

>> Power est.  Events/s    Category       Description
>>    8.18 W    100.0%      Device         USB device: Yubico Yubikey II (Yubico)
>>    8.13 W    100.0%      Device         USB device: Integrated Camera (Chicony Electronics Co., Ltd.)

having a USB device active also generally makes the CPU/etc use much more power,
which Powertop "charges" to the USB device in question.

8 Watts sounds entirely reasonable for that to me...

Re: [Powertop] USB devices get wrong power use values

[Marc MERLIN]

[-- Attachment #1: Type: text/plain, Size: 1280 bytes --]

On Sun, Oct 13, 2013 at 08:33:51AM -0700, Arjan van de Ven wrote:
> On 9/22/2013 4:54 PM, Marc MERLIN wrote:
> >So, I asked on the linux-usb list, and one thing I quickly found out is that the
> >Watt value given by powertop 2.4 is way off for USB devices:
> 
> what makes you think the value is off?
> 
> >>Power est.  Events/s    Category       Description
> >>   8.18 W    100.0%      Device         USB device: Yubico Yubikey II (Yubico)
> >>   8.13 W    100.0%      Device         USB device: Integrated Camera (Chicony Electronics Co., Ltd.)
> 
> having a USB device active also generally makes the CPU/etc use much more power,
> which Powertop "charges" to the USB device in question.
> 
> 8 Watts sounds entirely reasonable for that to me...

Aah, if the CPU gets involved, then yes, it's a bit more possible.
So, a simple USB keyboard that's unused can keep one core awake and burn 8
watts if something else goes wrong in the kernel to cause that?

Thanks,
Marc
-- 
"A mouse is a device used to point at the xterm you want to type in" - A.S.R.
Microsoft is to operating systems ....
                                      .... what McDonalds is to gourmet cooking
Home page: http://marc.merlins.org/                         | PGP 1024R/763BE901

Re: [Powertop] USB devices get wrong power use values

[Arjan van de Ven]

[-- Attachment #1: Type: text/plain, Size: 793 bytes --]

On 10/13/2013 9:22 AM, Marc MERLIN wrote:

>> having a USB device active also generally makes the CPU/etc use much more power,
>> which Powertop "charges" to the USB device in question.
>>
>> 8 Watts sounds entirely reasonable for that to me...
>
> Aah, if the CPU gets involved, then yes, it's a bit more possible.
> So, a simple USB keyboard that's unused can keep one core awake and burn 8
> watts if something else goes wrong in the kernel to cause that?

a bad USB keyboard can keep the processor out of deep package C states..
not a lot the software can do about that...
but yet it can be THAT much power delta.

(we've seen for example on some servers that a cheap keyboard makes > 15 Watts difference, but measured
at the wall socket, so after the power supply/etc)

Re: [Powertop] USB devices get wrong power use values

[Marc MERLIN]

[-- Attachment #1: Type: text/plain, Size: 1720 bytes --]

On Sun, Oct 13, 2013 at 11:48:25AM -0700, Arjan van de Ven wrote:
> On 10/13/2013 9:22 AM, Marc MERLIN wrote:
> 
> >>having a USB device active also generally makes the CPU/etc use much more power,
> >>which Powertop "charges" to the USB device in question.
> >>
> >>8 Watts sounds entirely reasonable for that to me...
> >
> >Aah, if the CPU gets involved, then yes, it's a bit more possible.
> >So, a simple USB keyboard that's unused can keep one core awake and burn 8
> >watts if something else goes wrong in the kernel to cause that?
> 
> a bad USB keyboard can keep the processor out of deep package C states..
> not a lot the software can do about that...
> but yet it can be THAT much power delta.
> 
> (we've seen for example on some servers that a cheap keyboard makes > 15 Watts difference, but measured
> at the wall socket, so after the power supply/etc)

So note that it's actually a very small USB widget that behaves as a
keyboard (it's an OTP token).
I've only had that OTP token show as 7W once, and it was suspicious that
it was at the exact same time as my built in camera.

So from what you're staying, you're pretty sure that the measuring code
was correct and that somehow the USB system was wedged or that power
management suspend failed in a way that both devices were taking 14W
together, more than what the entire laptop uses when on batteries with
CPUs mostly asleep?

Marc
-- 
"A mouse is a device used to point at the xterm you want to type in" - A.S.R.
Microsoft is to operating systems ....
                                      .... what McDonalds is to gourmet cooking
Home page: http://marc.merlins.org/                         | PGP 1024R/763BE901

Re: [Powertop] USB devices get wrong power use values

[Arjan van de Ven]

[-- Attachment #1: Type: text/plain, Size: 774 bytes --]

>
> So note that it's actually a very small USB widget that behaves as a
> keyboard (it's an OTP token).
> I've only had that OTP token show as 7W once, and it was suspicious that
> it was at the exact same time as my built in camera.
>
> So from what you're staying, you're pretty sure that the measuring code
> was correct and that somehow the USB system was wedged or that power
> management suspend failed in a way that both devices were taking 14W
> together, more than what the entire laptop uses when on batteries with
> CPUs mostly asleep?

well powertop cannot measure, only model these things
if you go WAAAY outside the current model (by adding devices never seen before)
the accuracy will go up over time, but initially might be off by a bit.

Re: [Powertop] USB devices get wrong power use values

[Marc MERLIN]

[-- Attachment #1: Type: text/plain, Size: 6506 bytes --]

On Sun, Oct 13, 2013 at 09:22:12AM -0700, Marc MERLIN wrote:
> On Sun, Oct 13, 2013 at 08:33:51AM -0700, Arjan van de Ven wrote:
> > On 9/22/2013 4:54 PM, Marc MERLIN wrote:
> > >So, I asked on the linux-usb list, and one thing I quickly found out is that the
> > >Watt value given by powertop 2.4 is way off for USB devices:
> > 
> > what makes you think the value is off?
> > 
> > >>Power est.  Events/s    Category       Description
> > >>   8.18 W    100.0%      Device         USB device: Yubico Yubikey II (Yubico)
> > >>   8.13 W    100.0%      Device         USB device: Integrated Camera (Chicony Electronics Co., Ltd.)
> > 
> > having a USB device active also generally makes the CPU/etc use much more power,
> > which Powertop "charges" to the USB device in question.
> > 
> > 8 Watts sounds entirely reasonable for that to me...
> 
> Aah, if the CPU gets involved, then yes, it's a bit more possible.
> So, a simple USB keyboard that's unused can keep one core awake and burn 8
> watts if something else goes wrong in the kernel to cause that?

So, I just upgraded to 3.11.5, and see things like this below. Can alsa sound
really take 10W when I'm batteries?
Let me put it another way, it just cannot be correct. Full power use is 15W and
if you removed 10W for audio (I'm sure that's wrong) and 3W for wifi (that may be correct)
it means my quad core laptop is running at 2W with the screen on?
Not possible at all.

So there has to be a bug in powertop 2.4 USB device power calculation.

For what it's worth, I got the data below right after running
powertop --calibrate and letting it go through a full cycle.

----------------------------------------------------------------------------
The battery reports a discharge rate of 15.1 W
The estimated remaining time is 4 hours, 53 minutes

Summary: 572.4 wakeups/second,  18.3 GPU ops/seconds, 0.0 VFS ops/sec and 8.3% CPU use

Power est.              Usage       Events/s    Category       Description
  10.5 W    100.0%                      Device         Audio codec alsa:hwC0D0: thinkpad (Realtek)
  3.10 W      7.0 pkts/s                Device         Network interface: wlan0 (iwlwifi)
  772 mW     25.1 ms/s     208.4        Process        /usr/bin/enlightenment
 13.4 mW      1.2 ms/s       8.5        Process        xfce4-terminal -T window10 --role=window10 --tab
(...)
----------------------------------------------------------------------------

Calibrate results if that helps:
gandalfthegreat:~# powertop --calibrate
Loaded 750 prior measurements
RAPL device for cpu 0
RAPL device for cpu 0
RAPL device for cpu 0
Starting PowerTOP power estimate calibration 
Calibrating idle
Calibrating: disk usage 
Calibrating backlight
.... device /sys/class/backlight/intel_backlight/brightness 
.... device /sys/class/backlight/acpi_video0/brightness 
Calibrating idle
Calibrating: CPU usage on 1 threads
Calibrating: CPU usage on 4 threads
Calibrating: CPU wakeup power consumption
Calibrating: CPU wakeup power consumption
Calibrating: CPU wakeup power consumption
Calibrating USB devices
.... device /sys/bus/usb/devices/3-1/power/control 
.... device /sys/bus/usb/devices/4-1/power/control 
.... device /sys/bus/usb/devices/usb1/power/control 
.... device /sys/bus/usb/devices/usb2/power/control 
.... device /sys/bus/usb/devices/usb3/power/control 
.... device /sys/bus/usb/devices/usb4/power/control 
.... device /sys/bus/usb/devices/3-1.2/power/control 
.... device /sys/bus/usb/devices/3-1.6/power/control 
Calibrating radio devices
.... device /sys/class/rfkill/rfkill0/soft 
.... device /sys/class/rfkill/rfkill1/soft 
.... device /sys/class/rfkill/rfkill0/soft 
.... device /sys/class/rfkill/rfkill1/soft 
Finishing PowerTOP power estimate calibration 
Parameters after calibration:


Parameter state 
----------------------------------
Value		Name
 2.23		alsa-codec-power (2)
 0.37		backlight (3)
 3.51		backlight-boost-100 (4)
 2.52		backlight-boost-40 (5)
 6.65		backlight-boost-80 (6)
 0.00		backlight-power (7)
 0.21		base power (8)
 0.00		cpu-consumption (9)
 0.00		cpu-wakeups (10)
 0.00		disk-operations (11)
 0.00		disk-operations-hard (12)
 0.00		eth0-link-100 (13)
10.93		eth0-link-1000 (14)
 0.00		eth0-link-high (15)
 0.00		eth0-packets (16)
 0.00		eth0-powerunsave (17)
 0.00		eth0-up (18)
 0.20		gpu-operations (19)
 0.00		radio:hci0 (20)
 0.00		radio:phy0 (21)
 0.31		radio:tpacpi_bluetooth_sw (22)
 0.00		runtime-0000:00:00.0 (23)
 0.00		runtime-0000:00:02.0 (24)
 0.00		runtime-0000:00:14.0 (25)
 0.00		runtime-0000:00:16.0 (26)
 0.00		runtime-0000:00:16.3 (27)
 0.00		runtime-0000:00:19.0 (28)
 0.00		runtime-0000:00:1a.0 (29)
 0.00		runtime-0000:00:1b.0 (30)
 0.00		runtime-0000:00:1c.0 (31)
 0.00		runtime-0000:00:1c.1 (32)
 0.00		runtime-0000:00:1c.2 (33)
 0.00		runtime-0000:00:1d.0 (34)
 0.00		runtime-0000:00:1f.0 (35)
 0.00		runtime-0000:00:1f.2 (36)
 0.00		runtime-0000:00:1f.3 (37)
 0.00		runtime-0000:02:00.0 (38)
 0.00		runtime-0000:03:00.0 (39)
 0.00		thinkpad-fan (40)
 0.00		thinkpad-fan-cub (41)
 0.00		thinkpad-fan-sqr (42)
 0.00		thinkpad-light (43)
 0.00		usb-device-0403-6001 (44)
 0.00		usb-device-04e8-6860 (45)
 0.00		usb-device-04e8-6864 (46)
 2.58		usb-device-04f2-b2ea (47)
 0.00		usb-device-058f-6387 (48)
 0.00		usb-device-091e-0003 (49)
 1.34		usb-device-0a5c-21e6 (50)
 9.82		usb-device-1050-0010 (51)
 0.00		usb-device-18d1-4ee2 (52)
 0.00		usb-device-18d1-4ee4 (53)
 0.00		usb-device-1d6b-0002 (54)
 0.00		usb-device-1d6b-0003 (55)
 0.00		usb-device-8087-0024 (56)
 0.00		usb0-link-100 (57)
 0.00		usb0-link-1000 (58)
 0.00		usb0-link-high (59)
 0.00		usb0-packets (60)
 0.00		usb0-powerunsave (61)
 0.00		usb0-up (62)
 0.00		vboxnet0-link-100 (63)
 0.00		vboxnet0-link-1000 (64)
 0.00		vboxnet0-link-high (65)
 0.00		vboxnet0-packets (66)
 0.00		vboxnet0-powerunsave (67)
 0.00		vboxnet0-up (68)
 0.00		wlan0-link-100 (69)
 0.00		wlan0-link-1000 (70)
 0.00		wlan0-link-high (71)
 0.00		wlan0-packets (72)
 1.64		wlan0-powerunsave (73)
 0.00		wlan0-up (74)
 0.00		xwakes (75)

Score:    2.6  (104691.5)
Guess:   13.2
Actual:  13.6
----------------------------------


-- 
"A mouse is a device used to point at the xterm you want to type in" - A.S.R.
Microsoft is to operating systems ....
                                      .... what McDonalds is to gourmet cooking
Home page: http://marc.merlins.org/  

Re: [Powertop] USB devices get wrong power use values

[Marc MERLIN]

[-- Attachment #1: Type: text/plain, Size: 4079 bytes --]

On Thu, Oct 17, 2013 at 05:01:54PM -0700, Marc MERLIN wrote:
> On Sun, Oct 13, 2013 at 09:22:12AM -0700, Marc MERLIN wrote:
> > On Sun, Oct 13, 2013 at 08:33:51AM -0700, Arjan van de Ven wrote:
> > > On 9/22/2013 4:54 PM, Marc MERLIN wrote:
> > > >So, I asked on the linux-usb list, and one thing I quickly found out is that the
> > > >Watt value given by powertop 2.4 is way off for USB devices:
> > > 
> > > what makes you think the value is off?
> > > 
> > > >>Power est.  Events/s    Category       Description
> > > >>   8.18 W    100.0%      Device         USB device: Yubico Yubikey II (Yubico)
> > > >>   8.13 W    100.0%      Device         USB device: Integrated Camera (Chicony Electronics Co., Ltd.)
> > > 
> > > having a USB device active also generally makes the CPU/etc use much more power,
> > > which Powertop "charges" to the USB device in question.
> > > 
> > > 8 Watts sounds entirely reasonable for that to me...
> > 
> > Aah, if the CPU gets involved, then yes, it's a bit more possible.
> > So, a simple USB keyboard that's unused can keep one core awake and burn 8
> > watts if something else goes wrong in the kernel to cause that?
> 
> So, I just upgraded to 3.11.5, and see things like this below. Can alsa sound
> really take 10W when I'm batteries?
> Let me put it another way, it just cannot be correct. Full power use is 15W and
> if you removed 10W for audio (I'm sure that's wrong) and 3W for wifi (that may be correct)
> it means my quad core laptop is running at 2W with the screen on?
> Not possible at all.

> So there has to be a bug in powertop 2.4 USB device power calculation.
> 
> For what it's worth, I got the data below right after running
> powertop --calibrate and letting it go through a full cycle.
> 
> ----------------------------------------------------------------------------
> The battery reports a discharge rate of 15.1 W
> The estimated remaining time is 4 hours, 53 minutes
> 
> Summary: 572.4 wakeups/second,  18.3 GPU ops/seconds, 0.0 VFS ops/sec and 8.3% CPU use
> 
> Power est.              Usage       Events/s    Category       Description
>   10.5 W    100.0%                      Device         Audio codec alsa:hwC0D0: thinkpad (Realtek)
>   3.10 W      7.0 pkts/s                Device         Network interface: wlan0 (iwlwifi)
>   772 mW     25.1 ms/s     208.4        Process        /usr/bin/enlightenment
>  13.4 mW      1.2 ms/s       8.5        Process        xfce4-terminal -T window10 --role=window10 --tab
> (...)
> ----------------------------------------------------------------------------

Just after I wrote this, I'm now seeing my yubikey back to 10W when it was
alsa holding the slot earlier.
 
> ----------------------------------------------------------------------------
The battery reports a discharge rate of 21.3 W
The estimated remaining time is 3 hours, 0 minutes

Summary: 2685.1 wakeups/second,  149.8 GPU ops/seconds, 0.0 VFS ops/sec and 31.2% CPU use

Power est.              Usage       Events/s    Category       Description
  10.4 W    100.0%                      Device         USB device: Yubico Yubikey II (Yubico)
  5.50 W     13.0%                      Device         Display backlight
  1.51 W     17.4 pkts/s                Device         Network interface: wlan0 (iwlwifi)
  566 mW     85.8 ms/s     1520.4       Process        /usr/bin/enlightenment
  109 mW    100.0%                      Device         Audio codec alsa:hwC0D0: thinkpad (Realtek)
 43.2 mW      5.2 ms/s      72.9        Process        xfce4-terminal -T window5 --role=window5
 25.9 mW      0.8 ms/s      14.8        Process        xfce4-terminal -T window9 --role=window9 --tab
> ----------------------------------------------------------------------------

I hope this is useful.

Marc
-- 
"A mouse is a device used to point at the xterm you want to type in" - A.S.R.
Microsoft is to operating systems ....
                                      .... what McDonalds is to gourmet cooking
Home page: http://marc.merlins.org/