On Thu, Mar 21, 2019 at 03:47:53PM +0000, Charles Keepax wrote: > On Thu, Mar 21, 2019 at 06:14:23AM -0700, Guenter Roeck wrote: > > On 3/21/19 4:59 AM, Charles Keepax wrote: > > >On Wed, Mar 20, 2019 at 09:40:10AM -0700, Guenter Roeck wrote: > > >>On Wed, Mar 20, 2019 at 02:58:18PM +0000, Charles Keepax wrote: > > >>>From: Lucas Tanure <tanureal@xxxxxxxxxxxxxxxxxxxxx> > > >>Is this an ieee754 floating point format ? Please add a comment stating it. > > >>Also, if the format supports it, please check for NaN. If the HW guarantees > > >>to never return NaN, please add a respective comment. > > >> > > >>How likely is it that the values overflow ? A precision multipler of > > >>1,000,000,000 means that numbers will overflow quite easily. And does > > >>the the hardware really report voltages and currents in pico-units, > > >>and are temperatures really reported in micro-degrees C ? > > >> > > > > > >I believe the hardware can't return NaN. But will check that, > > >what ranges are possible and add some overflow checking. > > > > > > > How about the units ? Maybe add a note explaining what the HW actually returns. > > > > Yup can do. > > > >>Overall it might make sense to reconfigure the hardware into continuous > > >>measurement mode and get rid of the delays (if that can be configured - > > >>the user guide isn't detailed enough to be able to determine for sure). > > >>After all, it is quite unlikely that the board will be used in an > > >>environment where the power savings would be worth the inconvenience > > >>of having to wait more than two seconds for a set of measurement values > > >>(adding all the current and temperature delays up). > > >> > > > > > >Yeah the hardware is quite slow and regrettably doesn't have a > > >continuous measurement mode. We could potentially add a thread to > > >poll them but mostly the usage for this data is just taking power > > >measurements of various audio use-cases so the large delay isn't > > >a huge problem and not sure it warrents the additional > > >complexity. > > > > > >>>+ msleep(nsamples); > > >>>+ > > >> > > >>This needs some explanation how nsamples translates into millisecond waits, > > >>especially since that wait can add up significantly (reading the temperature > > >>will take forever, as will reading all currents). > > >> > > >>>+ ret = regmap_read_poll_timeout(regmap, LOCHNAGAR2_IMON_CTRL3, val, > > >>>+ val & LOCHNAGAR2_IMON_DONE_MASK, > > >>>+ 5000, 2000000); > > >> > > >>Can that indeed take another two seconds on top of the sleep above ? > > >> > > > > > >It does about about 1.5-2 seconds I think last time I checked. > > > > > >Essentially the hardware will average a number of readings and > > >return that. The number of readings given in the driver is what > > >the hardware guys recommended for each, that said we could > > >potentially make it configurable if that helps at all? Also > > >I could get the msleep a bit closer to the actual runtime, > > >just need to check how linear the time is with the number of > > >samples take (I assume very). > > > > > > > Brr. Just add a note explaining that the long times are indeed intentional, > > and that the HW takes that long. > > > > From some more detailed discussions with the hardware guys > (shame on me for following the documentation) it seems > there isn't really much need for the temp to take so many > measurements. The current one makes sense as the hardware > actually does some analogue averaging so it should take account > of spikes in the current draw from the part under test. But > the temp should really only change slowly. I have also managed > to refine the estimates for the time taken by the measurements > so will factor that lot into the v2 as well. > Ok, that makes more sense. > Thanks very much for the review, will probably take me a few more Mu pleasure. And many thanks for the new macro - that is really very useful. > days to beat the last few clarifications out of the hardware guys > then I will fire up a v2. > Take your time. It is a few weeks until the next commit window, after all. Thanks, Guenter