Hi Alexandre,
Le mar., avril 19 2022 at 21:35:35 +0200, Alexandre Belloni
<alexandre.belloni@xxxxxxxxxxx> a écrit :
> On 18/04/2022 19:49:31+0100, Paul Cercueil wrote:
> > On power loss, reading the RTC value would fail as the
scratchpad
> > lost
> > its magic value, until the hardware clock was set once again.
> >
> > To avoid that, reset the RTC value to Epoch in the probe if we
> > detect
> > that the scratchpad lost its magic value.
> >
> > Signed-off-by: Paul Cercueil <paul@xxxxxxxxxxxxxxx>
> > ---
> > drivers/rtc/rtc-jz4740.c | 24 +++++++++++++++++++++---
> > 1 file changed, 21 insertions(+), 3 deletions(-)
> >
> > diff --git a/drivers/rtc/rtc-jz4740.c
b/drivers/rtc/rtc-jz4740.c
> > index 119baf168b32..aac5f68bf626 100644
> > --- a/drivers/rtc/rtc-jz4740.c
> > +++ b/drivers/rtc/rtc-jz4740.c
> > @@ -42,6 +42,9 @@
> > /* Magic value to enable writes on jz4780 */
> > #define JZ_RTC_WENR_MAGIC 0xA55A
> >
> > +/* Value written to the scratchpad to detect power losses */
> > +#define JZ_RTC_SCRATCHPAD_MAGIC 0x12345678
> > +
> > #define JZ_RTC_WAKEUP_FILTER_MASK 0x0000FFE0
> > #define JZ_RTC_RESET_COUNTER_MASK 0x00000FE0
> >
> > @@ -134,10 +137,11 @@ static int
jz4740_rtc_ctrl_set_bits(struct
> > jz4740_rtc *rtc, uint32_t mask,
> > static int jz4740_rtc_read_time(struct device *dev, struct
> > rtc_time *time)
> > {
> > struct jz4740_rtc *rtc = dev_get_drvdata(dev);
> > - uint32_t secs, secs2;
> > + uint32_t secs, secs2, magic;
> > int timeout = 5;
> >
> > - if (jz4740_rtc_reg_read(rtc, JZ_REG_RTC_SCRATCHPAD) !=
0x12345678)
> > + magic = jz4740_rtc_reg_read(rtc, JZ_REG_RTC_SCRATCHPAD);
> > + if (magic != JZ_RTC_SCRATCHPAD_MAGIC)
> > return -EINVAL;
> >
> > /* If the seconds register is read while it is updated, it
can
> > contain a
> > @@ -169,7 +173,8 @@ static int jz4740_rtc_set_time(struct
device
> > *dev, struct rtc_time *time)
> > if (ret)
> > return ret;
> >
> > - return jz4740_rtc_reg_write(rtc, JZ_REG_RTC_SCRATCHPAD,
> > 0x12345678);
> > + return jz4740_rtc_reg_write(rtc, JZ_REG_RTC_SCRATCHPAD,
> > + JZ_RTC_SCRATCHPAD_MAGIC);
> > }
> >
> > static int jz4740_rtc_read_alarm(struct device *dev, struct
> > rtc_wkalrm *alrm)
> > @@ -307,6 +312,7 @@ static int jz4740_rtc_probe(struct
> > platform_device *pdev)
> > struct jz4740_rtc *rtc;
> > unsigned long rate;
> > struct clk *clk;
> > + uint32_t magic;
> > int ret, irq;
> >
> > rtc = devm_kzalloc(dev, sizeof(*rtc), GFP_KERNEL);
> > @@ -369,6 +375,18 @@ static int jz4740_rtc_probe(struct
> > platform_device *pdev)
> > /* Each 1 Hz pulse should happen after (rate) ticks */
> > jz4740_rtc_reg_write(rtc, JZ_REG_RTC_REGULATOR, rate - 1);
> >
> > + magic = jz4740_rtc_reg_read(rtc, JZ_REG_RTC_SCRATCHPAD);
> > + if (magic != JZ_RTC_SCRATCHPAD_MAGIC) {
> > + /*
> > + * If the scratchpad doesn't hold our magic value, then a
> > + * power loss occurred. Reset to Epoch.
> > + */
> > + struct rtc_time time;
> > +
> > + rtc_time64_to_tm(0, &time);
> > + jz4740_rtc_set_time(dev, &time);
>
> Don't do that, this defeats the purpose of detecting when the
power is
> lost. Returning a known bogus time is the worst thing you can do
here.
So what is the best thing to do then?
