Re: [PATCH] drivers: mfd: da9063: Add restart notifier implementation

[David Abdurachmanov <david.abdurachmanov@gmail.com>]

On Wed, Sep 29, 2021 at 4:36 PM Adam Thomson
<Adam.Thomson.Opensource@diasemi.com> wrote:
>
> On 24 September 2021 17:17, Alexandre Ghiti wrote:
>
> > > > +static int da9063_restart_notify(struct notifier_block *this,
> > > > +                              unsigned long mode, void *cmd)
> > > > +{
> > > > +     struct da9063 *da9063 = container_of(this, struct da9063,
> > > > restart_handler);
> > > > +
> > > > +     regmap_write(da9063->regmap, DA9063_REG_PAGE_CON, 0x00);
> > > > +     regmap_write(da9063->regmap, DA9063_REG_CONTROL_F, 0x04);
> > > > +     regmap_write(da9063->regmap, DA9063_REG_CONTROL_A, 0x68);
> > > > +
> > > > +     return NOTIFY_DONE;
> > > > +}
> > >
> > > I will talk with our HW team to clarify, but this sequence looks to be very
> > > specific to the needs of the platform in question which doesn't feel right to
> > > me. As was mentioned on another thread as well, the watchdog driver already
> > has
> > > a restart function to reset the device (and thus the system), so I don't believe
> > > we should have multiple of these.
> >
> > From the discussion that happened here
> > https://www.dialog-semiconductor.com/products/pmics?post_id=10052#tab-
> > support_tab_content,
> > it does not seem possible to use the watchdog on a chip whose OTP does
> > not set AUTOBOOT. But anyway, I'm looking forward to hearing from the
> > HW team :)
>
> So I've discussed this internally and so far it's not completely clear how the
> sequence you provided actually performs the reset as you suggest. It certainly
> doesn't look like it should, so maybe this relates to an external pin somehow
> triggering the restart in this particular scenario? I'd be interested to
> understand which event bits are set when the board does restart to understand
> what did actually trigger the boot-up.
>
> Regardless of this though, the consensus right now would be to use the RTC as a
> wake event to restart the platform. An alarm can be set for a couple of seconds
> into the future (or longer if required) and that would provide the event
> required to come up from powerdown/shutdown, in the absence of AUTOBOOT being
> set in OTP. I believe this would be the safest route to take in this case. You
> can then just use the SHUTDOWN bit on CONTROL_F to take down the board.

Today I was looking into OpenBSD DA9063 drivers and they might be
doing what you described for the reset.

dev/fdt/dapmic.c

[..]
241 void
242 dapmic_reset(void)
243 {
244     struct dapmic_softc *sc = dapmic_cd.cd_devs[0];
245     uint8_t reg;
246
247     /* Enable tick alarm wakeup with a one second interval. */
248     reg = dapmic_reg_read(sc, ALARM_MO);
249     reg &= ~ALARM_MO_TICK_TYPE;
250     reg |= ALARM_MO_TICK_WAKE;
251     dapmic_reg_write(sc, ALARM_MO, reg);
252
253     /* Enable tick function. */
254     reg = dapmic_reg_read(sc, ALARM_Y);
255     reg |= ALARM_Y_TICK_ON;
256     dapmic_reg_write(sc, ALARM_Y, reg);
257
258     /* Clear events such that we wake up again. */
259     dapmic_reg_write(sc, EVENT_A, dapmic_reg_read(sc, EVENT_A));
260     dapmic_reg_write(sc, CONTROL_F, CONTROL_F_SHUTDOWN);
261 }
[..]

>
> To reiterate, I believe this should be made a board specific quirk, rather than
> as part of the generic MFD core of DA9063, as the timings may vary for other
> platforms.

Agree. Currently it seems Linux drivers expect DA9063 boards to have
AUTOBOOT ON set in OTP, which is not the case for SiFive Unmatched
(thus issues with reset and WDT).

david

> _______________________________________________
> linux-riscv mailing list
> linux-riscv@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-riscv