Re: [PATCH v5 0/3] iio: adc: add support for the MAX34417 Four-Channel High Dynamic Range Power Accumulator
From: Jonathan Cameron
Date: Mon Nov 16 2026 - 22:23:08 EST
On Mon, 16 Nov 2026 10:34:30 +0100
Neil Armstrong <neil.armstrong@xxxxxxxxxx> wrote:
> On 11/14/26 19:46, Jonathan Cameron wrote:
> > On Mon, 9 Nov 2026 10:52:22 +0100
> > Neil Armstrong <neil.armstrong@xxxxxxxxxx> wrote:
> >
> >> On 11/8/26 00:09, Jonathan Cameron wrote:
> >>> On Wed, 4 Nov 2026 09:45:02 +0100
> >>> Neil Armstrong <neil.armstrong@xxxxxxxxxx> wrote:
> >>
> >> <snip>
> >>
> >>>> But for the other reports:
> >>>> - [Medium] iio: adc: max34417: shared accumulators clobbered on CAM power/energy read
> >>>>
> >>>> This is a tough one, today if either we read power or energy, we erase the accumulated
> >>>> power+count in the context, the context accumulator is only here to make sure we don't
> >>>> loose data from other channels, like if we had a single HW channel.
> >>>> This simulates this, if we had a single HW channel, at each UPDATE we would get back
> >>>> to 0 in hardware anyway if we did read the power or energy.
> >>>
> >>>
> >>> I'm only half understanding this. Let me see...
> >>>
> >>> I'd expect userspace to see accumulated values since we enabled the channel
> >>> rather than since last read. I think that's the final suggestion from
> >>> sashiko.
> >>
> >> Oh this changes quite a lot of things.
> >> How do we handle overflow in this case ? we disable the channel and enable again to reset the state ?
> >
> > Is it possible to accumulate into a sufficiently large type that overflow won't happen in a feasible
> > timescale? Anything else is going to be a but unpredictable for userspace if it
> > is expecting an accumulated number. The other things that accumulate in IIO tend to be
> > moderately small so it's never been an issue (things like step counters).
>
> The hardware can store up to 4.55h of data with it's 56bit with the default rate,
> we store in 64bit in software, so we can store up to 256 times so up to 48.53 days.
>
> It's possible to lower the SPS to 8, in this mode we can store up to 6211 days of data.
> I plan to add this mode afterwards.
>
> I'll switch to this and clear the overflow only if we disable the POWER channel.
Hmm. Could use dual 64 bits values and a retry loop for theoretical tearing.
Would then need custom code to report though which isn't ideal and
it might rather surprise userspace.
>
> Neil
>
> >
> > Jonathan
>