[PATCH v2 3/7] power: regulator-uclass: perform regulator setup inside uclass

[Svyatoslav Ryhel]

нд, 6 серп. 2023 р. о 19:21 Svyatoslav Ryhel <clamor95 at gmail.com> пише:
>
> сб, 5 серп. 2023 р. о 15:49 Jonas Karlman <jonas at kwiboo.se> пише:
> >
> > Hi,
> >
> > On 2023-07-21 07:34, Svyatoslav Ryhel wrote:
> > > чт, 20 лип. 2023 р. о 22:43 Simon Glass <sjg at chromium.org> пише:
> > >>
> > >> Hi Svyatoslav,
> > >>
> > >> On Thu, 20 Jul 2023 at 06:38, Svyatoslav Ryhel <clamor95 at gmail.com> wrote:
> > >>>
> > >>> Regulators initial setup was previously dependent on board call.
> > >>> To move from this behaviour were introduced two steps. First is
> > >>> that all individual regulators will be probed just after binding
> > >>
> > >> We must not probe devices automatically when bound. The i2c interface
> > >> may not be available, etc. Do a probe afterwards.
> > >>
> > >> Perhaps I am misunderstanding this, iwc please reword this commit message.
> > >
> > > After bound. If the regulator is a self-sufficient i2c device then it
> > > will be bound
> > > after i2c is available by code design so i2c interface should be
> > > available at that
> > > moment. At least led and gpio uclasses perform this for initial setup
> > > of devices.
> > >
> > > Platform regulators (aka fixed/gpio regulators) work perfectly fine. I
> > > have no i2c
> > > regulators to test deeply.
> > >
> > > As for now only one uclass is not compatible with this system - PMIC which has
> > > strong dependency between regulator and main mfd. This is why probing after
> > > bind for pmic regulators is disabled and pmic regulators are probed on pmic core
> > > post_probe.
> >
> > This sounds more like a possible bug of some dependency not being
> > probed in correct order or not leaving the device in a ready state
> > after probe.
> >
> > If pmic regulators are bind in pmic bind with the pmic as parent, then
> > at regulator probe the driver core ensure that pmic has probed before
> > regulator use.
>
> I have found why this occurs. With this patchset always/boot-on regulators
> are probed immediately after bind, since pmic regulators are bound on pmics
> bind it causes a situation when regulator is probed before pmic and pmics ops
> are not available. Moving pmic children bind to pmics probe fixed this issue.
>
I have made a mailing list reduction. This issue goes deeper, both pmic and its
regulator children work as expected with no issues there. I get error from i2c
driver directly that it fails on xfer. This function

https://github.com/clamor-s/u-boot/blob/master/drivers/i2c/tegra_i2c.c#L477

returns me EREMOTEIO. At the same time if the regulator is autoset on pmics
post_probe I do not have this error. Maybe you have any suggestions on how to
handle this situation?

Best regards,
Svyatoslav R.

> > This works perfect fine with the RK8xx I2C PMIC and its regulators.
> > Wich a call to device_get_supply_regulator(dev, "test-supply", &reg)
> > probe happens in i2c, pmic and regulator order.
> >
> > >
> > >>> which ensures that regulator pdata is filled and second is setting
> > >>> up regulator in post probe which enseres that correct regulator
> > >>> state according to pdata is reached.
> >
> > I think that the approach in this patch is trying to change too much all
> > at once.
> >
> > Have made some adjustments that I think should help with transition.
> > - Added a flag to protect regulator_autoset from being called more then
> >   once for each regulator, in a separate patch.
> > - Changed to only probe-after-bind on regulators marked as
> >   always-on/boot-on.
> >
> > And it works something like this:
> >
> > static int regulator_post_bind(struct udevice *dev)
> > {
> >         [...]
> >
> >         if (uc_pdata->always_on || uc_pdata->boot_on)
> >                 dev_or_flags(dev, DM_FLAG_PROBE_AFTER_BIND);
> > }
> >
> > static int regulator_post_probe(struct udevice *dev)
> > {
> >         ret = regulator_autoset(dev);
> > }
> >
> > With that any always-on/boot-on regulator would automatically probe and
> > autoset after bind, remaining would only probe and autoset if they are
> > used.
> >
> > This approach should also prevent having to change existing code that
> > may call autoset, and cleanup can be done in follow-up patches/series.
> >
> > Probe-after-bind and call to autoset could also be protected with a new
> > Kconfig to help with transition.
> >
> > See following for a working example using a new driver that depends
> > on that regulators have been autoset prior to regulator_get_value.
> > https://github.com/Kwiboo/u-boot-rockchip/compare/master...rk3568-io-domain
> >
> > Or the two commits separate:
> >
> > power: regulator: Only run autoset once for each regulator
> > https://github.com/Kwiboo/u-boot-rockchip/commit/05db4dbcb8f908b417ed5cae8a7a325c82112d75
> >
> > power: regulator: Perform regulator setup inside uclass
> > https://github.com/Kwiboo/u-boot-rockchip/commit/489bd5d2c1a2a33824eac4f2d54399ef5dff4fdf
> >
> > Regards,
> > Jonas
> >
> > >>>
> > >>> Signed-off-by: Svyatoslav Ryhel <clamor95 at gmail.com>
> > >>> ---
> > >>>  drivers/power/regulator/regulator-uclass.c | 212 ++++++---------------
> > >>>  include/power/regulator.h                  | 119 ------------
> > >>>  2 files changed, 62 insertions(+), 269 deletions(-)
> > >>
> > >> Regards,
> > >> SImon
> >