Re: [PATCH v2 01/10] mfd: intel_soc_pmic: Fix an error handling path in intel_soc_pmic_i2c_probe()

[Andy Shevchenko]

On Mon, Aug 1, 2022 at 11:52 AM Hans de Goede <hdegoede@xxxxxxxxxx> wrote:
> On 8/1/22 11:29, Andy Shevchenko wrote:
> > On Mon, Aug 1, 2022 at 11:14 AM Andy Shevchenko
> > <andy.shevchenko@xxxxxxxxx> wrote:
> >> On Mon, Aug 1, 2022 at 10:43 AM Hans de Goede <hdegoede@xxxxxxxxxx> wrote:
> >>> On 7/31/22 22:12, Andy Shevchenko wrote:

...

> >>> Note alternatively we could just move the pwm_add_table() to just before the "return 0",
> >>> there is no strict ordering between adding the mfd devices and the pwm_add_table()
> >>> (the pwm device only becomes available after the pwm-driver has bound to the mfd
> >>> instantiated platform device which happens later).
> >
> > Just to be sure... How is it guaranteed that that happens later?
>
> Ah you are right, it could happen immediately if the driver is builtin and
> has already registered (if the PWM driver is a module, as it is on Fedora,
> then the driver will only bind once the module is loaded).
>
> Regardless there are no ordering guarantees between the probe() function of
> intel_soc_pmic and the consumer of the PWM device, so the consumer must
> be prepared to deal with the lookup not being present yet when its probe()
> function runs (*).

Would be nice to have, but isn't it the issue with all lookup tables
so far, e.g. consumers of GPIO ones are also affected the very same
way?

> *) ATM this is actually an unsolved problem and this works only because the PMIC
> drivers are builtin and i915, which consumes the PWM for backlight control
> is a module. Swapping the order does not impact this.

Even so, I think we can't change order right now because the issue is
much broader.

-- 
With Best Regards,
Andy Shevchenko