Re: [PATCH INTERNAL v1 1/3] rtc: tps6594: add driver for TPS6594 PMIC RTC

[Alexandre Belloni <alexandre.belloni@xxxxxxxxxxx>]

On 13/03/2023 13:10:37+0100, Esteban Blanc wrote:
> On Mon Mar 13, 2023 at 12:01 PM CET, Alexandre Belloni wrote:
> > On 13/03/2023 10:18:45+0100, Esteban Blanc wrote:
> > > On Tue Mar 7, 2023 at 12:08 PM CET, Alexandre Belloni wrote:
> > > > On 24/02/2023 14:31:27+0100, Esteban Blanc wrote:
> > > > > +/*
> > > > > + * Gets current tps6594 RTC time and date parameters.
> > > > > + *
> > > > > + * The RTC's time/alarm representation is not what gmtime(3) requires
> > > > > + * Linux to use:
> > > > > + *
> > > > > + *  - Months are 1..12 vs Linux 0-11
> > > > > + *  - Years are 0..99 vs Linux 1900..N (we assume 21st century)
> > > > > + */
> > > >
> > > > I don't find this comment to be particularly useful.
> > > 
> > > Ok. I propose that I add 2 constants for the -1 and +100 in the month and year
> > > calculation. This way, without the comment the computation would be a
> > > bit more self explanatory.
> > > What do you think?
> >
> > I don't think this is necessary, keep -1 for the month and +100 for the
> > year, those are very common operations in the subsystem and don't really
> > need any explanation
> 
> Ok. I will just remove the comment then.
> 
> > > > > +static int tps6594_rtc_probe(struct platform_device *pdev)
> > > > > +{
> > > > > +   struct tps6594 *tps6594;
> > > > > +   struct tps6594_rtc *tps_rtc;
> > > > > +   int irq;
> > > > > +   int ret;
> > > > > +
> > > > > +   tps6594 = dev_get_drvdata(pdev->dev.parent);
> > > > > +
> > > > > +   tps_rtc = devm_kzalloc(&pdev->dev, sizeof(struct tps6594_rtc),
> > > > > +                          GFP_KERNEL);
> > > > > +   if (!tps_rtc)
> > > > > +           return -ENOMEM;
> > > > > +
> > > > > +   tps_rtc->rtc = devm_rtc_allocate_device(&pdev->dev);
> > > > > +   if (IS_ERR(tps_rtc->rtc))
> > > > > +           return PTR_ERR(tps_rtc->rtc);
> > > > > +
> > > > > +   /* Enable crystal oscillator */
> > > > > +   ret = regmap_set_bits(tps6594->regmap, TPS6594_REG_RTC_CTRL_2,
> > > > > +                         TPS6594_BIT_XTAL_EN);
> > > > > +   if (ret < 0)
> > > > > +           return ret;
> > > > > +
> > > > > +   /* Start rtc */
> > > > > +   ret = regmap_set_bits(tps6594->regmap, TPS6594_REG_RTC_CTRL_1,
> > > > > +                         TPS6594_BIT_STOP_RTC);
> > > > > +   if (ret < 0)
> > > > > +           return ret;
> > > >
> > > > Do that (XTAL_EN and clearing STOP) only once the time is known to be
> > > > set to a correct value so read_time doesn't have a chance to return a
> > > > bogus value.
> > > >
> > > 
> > > (...)
> > > 
> > > I understand your point, however I'm not sure of the canonical way to do
> > > this. Simply calling `tps6594_rtc_set_time` is enough?
> >
> > Yeah, let userspace set the time and start the rtc at that point.
> 
> The problem with that is we might have some RTCs that will just not be
> usable. We have boards with multiple TP6594 PMICs where only one of them
> has a crystal oscillator. The way to detect this is to start the RTC
> then checked if the STOP_RTC bit is still 0. By doing this in the probe,
> I'm able to not register an RTC device that doesn't work.
> 
> If I just start the RTC on the first call to `tps6594_rtc_set_time`, it
> will work for the RTC with the crystal and fails for all the others 
> 
> I can stop the RTC at the end of the probe, after the check to rule out
> unusable devices. If I add the check you proposed in
> `tps6594_rtc_read_time` it will fail until a successful call to
> `tps6594_rtc_set_time`. Would that be a suitable solution?
> 

That would work, yes


-- 
Alexandre Belloni, co-owner and COO, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com