[Gta04-owner] [Community] GTA04 power management [was: About cost and business models for community hardware]

Neil Brown neilb at suse.de
Thu Oct 8 08:55:28 CEST 2015


Christ van Willegen <cvwillegen at gmail.com> writes:

> On Sat, Oct 3, 2015 at 5:16 AM, Neil Brown <neilb at suse.de> wrote:
>> I run that several times and discard the outliers.
>>
>> With a 4.2 kernel at:
>>    http://git.neil.brown.name/?p=gta04.git;a=shortlog;h=refs/heads/4.2-gta04
>> I get numbers like:
>>
>> 46967 uA over 301 seconds
>> 46967 uA over 301 seconds
>> 47124 uA over 300 seconds
>> 47124 uA over 300 seconds
>> 47281 uA over 299 seconds
>> 47281 uA over 299 seconds
>> 47600 uA over 297 seconds
>>
>>
>> With the 3.7 kernel
>>  http://git.neil.brown.name/?p=gta04.git;a=shortlog;h=refs/heads/3.7-gta04
>> I get:
>>
>> 21420 uA over 300 seconds
>> 21420 uA over 300 seconds
>> 36420 uA over 300 seconds
>> 36420 uA over 300 seconds
>
> Is this with the same bootloader on both test trials?

Yes.

>
> Does setting pinmux, GPIOs, etc (anything that influences power usage)
> via 'a few' Linux kernel calls?
>
> If so, is it possible to do logging (via RS232, since that _is_
> available...) of the settings that are set via those kernel calls, so
> that the two can be compared?

pinmux settings can be read out of sysfs.  It's been a while, but I'm
fairly sure I looked into that.
There aren't very many gpios, but I'm pretty use I considered each of
them and found they were the same for both, or made no difference when
toggled.

>
> If you can switch back and forth between two kernel versions, and get
> higher and lower readings consistently, it's not the hardware, but the
> software that is incorrectly addressing the hardware. If logging is
> possible, it is possible to see the difference in setting (or order of
> setting) hardware parameters, and fix the software accordingly.
>
> Is this too short-sighted?

I have played some games like that: logging all writes to registers and
looking for differences.  There are an enormous number of differences
and any that I actually looked into were inconsequential.

Without some way to narrow the search space, it seems an intractable
problem.
The only thing I can think of would be a git-bisect.
But that would require patching each intermediate kernel to get it
working enough to boot/power-down-all-peripherals/test.
That would take multiple weeks for full-time work (at a guess) and I'm
not prepared to spend that - particularly as there is a rather large
transition from board-file to devicetree and the difference could be
hidden in there - not bisectable.

NeilBrown

>
> Christ van Willegen
> -- 
> 09 F9 11 02 9D 74 E3 5B D8 41 56 C5 63 56 88 C0
> _______________________________________________
> Gta04-owner mailing list
> Gta04-owner at goldelico.com
> http://lists.goldelico.com/mailman/listinfo.cgi/gta04-owner
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 818 bytes
Desc: not available
URL: <http://lists.goldelico.com/pipermail/gta04-owner/attachments/20151008/24435752/attachment.asc>


More information about the Gta04-owner mailing list