Re: [PATCH v4 11/11] platform/x86: int3472: Deal with probe ordering issues

From: Andy Shevchenko
Date: Mon Oct 25 2021 - 07:41:44 EST


On Mon, Oct 25, 2021 at 12:42 PM Hans de Goede <hdegoede@xxxxxxxxxx> wrote:
>
> The clk and regulator frameworks expect clk/regulator consumer-devices
> to have info about the consumed clks/regulators described in the device's
> fw_node.
>
> To work around this info missing from the ACPI tables on devices where
> the int3472 driver is used, the int3472 MFD-cell drivers attach info about
> consumers to the clks/regulators when registering these.
>
> This causes problems with the probe ordering wrt drivers for consumers
> of these clks/regulators. Since the lookups are only registered when the
> provider-driver binds, trying to get these clks/regulators before then
> results in a -ENOENT error for clks and a dummy regulator for regulators.
>
> All the sensor ACPI fw-nodes have a _DEP dependency on the INT3472 ACPI
> fw-node, so to work around these probe ordering issues the ACPI core /
> i2c-code does not instantiate the I2C-clients for any ACPI devices
> which have a _DEP dependency on an INT3472 ACPI device until all
> _DEP-s are met.
>
> This relies on acpi_dev_clear_dependencies() getting called by the driver
> for the _DEP-s when they are ready, add a acpi_dev_clear_dependencies()
> call to the discrete.c probe code.
>
> In the tps68470 case calling acpi_dev_clear_dependencies() is already done
> by the acpi_gpiochip_add() call done by the driver for the GPIO MFD cell
> (The GPIO cell is deliberately the last cell created to make sure the
> clk + regulator cells are already instantiated when this happens).
>
> However for proper probe ordering, the clk/regulator cells must not just
> be instantiated the must be fully ready (the clks + regulators must be
> registered with their subsystems).
>
> Add MODULE_SOFTDEP dependencies for the clk and regulator drivers for
> the instantiated MFD-cells so that these are loaded before us and so
> that they bind immediately when the platform-devs are instantiated.

Just a side note: MODULE_SOFTDEP() won't work in some (special?) cases
when module tools are limited in functionality (e.g. busybox
implementation as of today).

--
With Best Regards,
Andy Shevchenko