On Sun, 31 Jul 2022 21:51:55 +0300 Matti Lehtimäki <matti.lehtimaki@xxxxxxxxx> wrote: > On 31.7.2022 19.00, Jonathan Cameron wrote: > > On Sun, 24 Jul 2022 19:43:15 +0300 > > Matti Lehtimäki <matti.lehtimaki@xxxxxxxxx> wrote: > > > >> Some sensors do not always start fast enough to read a valid ID from > >> registers at first attempt. Let's retry at most 3 times with short sleep > >> in between to fix random timing issues. > >> > >> Signed-off-by: Matti Lehtimäki <matti.lehtimaki@xxxxxxxxx> > > Hi Matti, > > > > My gut feeling is this isn't in a fast path, so why not just wait > > for whatever the documented power up time of the sensor is? > > > > I'd expect to see a sleep in st_sensors_power_enable() if one is > > required. > > In the specification for the sensor (lis2hh12) I have on my device I > found that the maximum boot time of the sensor (starting from Vdd power > on) is defined as 20 ms. Not sure if the other sensors supported by the > driver have different values but based on checking a couple of > specifications I didn't find any bigger values so far. > > >> + msleep(20); > > How do we know 60msecs is long enough for all sensors? > > Based on the specification for the sensor I have and also driver used in > Android kernel for my device (it uses a 3 x 20 ms loop) I think 20 ms is > a good value but to be sure a slightly longer might make sense. As > suggested in the other review comment by changing the regmap_read to > regmap_read_poll_timeout the function doesn't always need to wait at > least 20 ms in case first read doesn't provide the correct value, if a > suitable shorter poll interval is used (something like 1-10 ms). > > However testing on my device has shown that I still need to have a loop > or at least a retry possibility because I have noticed a rare random > read error (-6, happens after some time not at first read) when reading > the id from the hardware. This could be due to for example internal > init failure of the sensor chip causing an internal reset. Because of > this read error regmap_read_poll_timeout returns with an error and > without retrying to read the id the sensor probe fails. Nasty. If you can get a confirmation that it's a possible failure on startup from the manufacturer then I'd be happier with that justification to retry rather than just sleep for say 30msec after power on. Jonathan > > -Matti
