On poniedziałek, 13 września 2021 15:02:34 CEST Krzysztof Kozlowski wrote:
> On 12/09/2021 22:54, Sebastian Krzyszkowiak wrote:
> > The gauge requires us to clear the status bits manually for some alerts
> > to be properly dismissed. Previously the IRQ was configured to react only
> > on falling edge, which wasn't technically correct (the ALRT line is active
> > low), but it had a happy side-effect of preventing interrupt storms
> > on uncleared alerts from happening.
> > 
> > Fixes: 7fbf6b731bca ("power: supply: max17042: Do not enforce (incorrect)
> > interrupt trigger type") Cc: <stable@vger.kernel.org>
> > Signed-off-by: Sebastian Krzyszkowiak <sebastian.krzyszkowiak@puri.sm>
> > ---
> > 
> >  drivers/power/supply/max17042_battery.c | 3 +++
> >  1 file changed, 3 insertions(+)
> > 
> > diff --git a/drivers/power/supply/max17042_battery.c
> > b/drivers/power/supply/max17042_battery.c index
> > 8dffae76b6a3..c53980c8432a 100644
> > --- a/drivers/power/supply/max17042_battery.c
> > +++ b/drivers/power/supply/max17042_battery.c
> > @@ -876,6 +876,9 @@ static irqreturn_t max17042_thread_handler(int id,
> > void *dev)> 
> >  		max17042_set_soc_threshold(chip, 1);
> >  	
> >  	}
> > 
> > +	regmap_clear_bits(chip->regmap, MAX17042_STATUS,
> > +			  0xFFFF & ~(STATUS_POR_BIT | 
STATUS_BST_BIT));
> > +
> 
> Are you sure that this was the reason of interrupt storm? Not incorrect
> SoC value (read from register for ModelGauge m3 while not configuring
> fuel gauge model).

Yes, I am sure. I have observed this on a fully configured max17055 with 
ModelGauge m5. It also makes sense to me based on what I read in the code and 
datasheets.

There were two kinds of storms - the short ones happening on each SOC change 
caused by SOC threshold alerts set by max17042_set_soc_threshold which 
eventually got cleared by reconfiguring the thresholds; and a huge one 
happening when SOC got down to 0% that did not get away until the battery got 
charged to at least 1% at which point the thresholds got reconfigured again 
(which is how I noticed the underflow fixed by the second patch).

Besides, I also have patches for configuring m5 gauge via DT that I'll send 
once I clean them up.

> You should only clear bits which you are awaken for... Have in mind that
> in DT-configuration the fuel gauge is most likely broken by missing
> configuration. With alert enabled, several other config fields should be
> cleared.

I have checked all the bits in the Status register and aside of Bst, POR and 
bunch of "don't-care" bits they're all alert indicators that we either handle 
explicitly in the interrupt handler (Smn/Smx) or implicitly via 
power_supply_changed (Imn/Imx, Vmn/Vmx, Tmn/Tmx, dSOCi, Bi/Br). The driver 
unconditionally enables alerts for SOC thresholds and all the rest stays 
effectively disabled at POR; however, a bootloader or firmware may configure it 
differently, which may be wanted for things like resuming from suspend when a 
bad condition happens. Therefore we need to clear all the bits anyway and I'm 
not sure whether iterating through them in a "if set then clear" loop gains us 
anything aside of additional lines of code.

> Best regards,
> Krzysztof

Cheers,
Sebastian