[Gta04-owner] Tracking down power 'leak'

Dr. H. Nikolaus Schaller hns at goldelico.com
Wed Sep 4 10:28:59 CEST 2013


Am 29.08.2013 um 07:55 schrieb Christ van Willegen:

> Hi all,
> Yesterday, I discussed our power 'leak' problem with an old EE.
> He suggested putting a small cable between the battery and the battery
> leads in order to put in an Ampere meter to study battery usage in
> more detail than what could be done by measuring the charge_now values
> before and after suspend. This would give an indication of power
> spikes (coming from the GSM modem) and things like that.

Yes. I had done similar setup a while ago to check if the modem is
really shutting down (yes, I could verify the 310 mA from the data sheets)
but did not find too much about the kernel.

Maybe with the new knowledge collected since then, I think it is worth
to repeat.

The main problem is the big variation range. It goes from uA in poweroff
and 3-10 mA in suspend to 2 A in full power mode. Typical Amperemeters
have a 200 mA and a 2A range - where you must replug the probes :(

Instead of an Amperemeter it would be better to insert a 0.1 Ohm shunt
resistor in the - (GND) line of the battery and connect an Oscilloscope.

> Also, would it be possible to build a minimal system that has either
> USB or serial support (I don't have a serial cable, so I can't do this
> at home...) and then connect to it and manually (using a simple C
> program tugging registers left and right) toggling all the hardware to
> the lowest power state possible. Then. when we are reasonably sure
> that everything is off, we can measure power usage to have a baseline
> figure.

Hm. Unfortunately the OMAP is so complex that there is no minimal
system allowing for this approach. The closest thing is to use U-Boot
(which is a single-user, no built-in power management OS). But
there we don't have protocol support to enable/disable some peripherals.

> In the real system, is there a way to catch the (kernel) calls that
> turn off various devices so that they can be logged somewhere? That
> would enable finding out which devices are still enabled in some way.
> No solutions in this mail, only questions, but perhaps they can
> inspire others to new insights?

Christoph has done yet another approach a while ago (I think he did report
it somewhere): 1. send the device to suspend, 2. use a very sensitive
infrared camera to detect those chips that still dissipate some mW.

If I remember correctly it did point out only CPU+RAM.


More information about the Gta04-owner mailing list