Re: [PATCH net v2 1/2] net: dsa: mt7530: fix NULL dereference on unbind of MT7531 and MT7621

[Aleksei Sviridkin <f@lex.la>]

> Did you consider using devm_regulator_get_enable()?

No. Looking at it now, it is the better shape, but not for this patch.

Two things stop it from being a swap. The driver sets the voltages before
it enables, 1.0 V on core and 3.3 V on io, and get_enable returns an int,
so no handle is left for regulator_set_voltage(). On the BPI-R2, which is
also the example in the binding, both supplies are MT6323 regulators with
a range, vpa 0.5 to 3.65 V and vemc3v3 3.0 to 3.3 V, so those calls are
not decoration. The other is placement: the get is in probe and the enable
is in mt7530_setup(), which DSA calls much later and which resets the chip
on the next lines, so get_enable would bring the rails up at a different
point.

It is still worth doing, for a reason this patch deliberately leaves
alone. mt7530_remove() disables the supplies before
mt7530_remove_common() calls dsa_unregister_switch(), which still writes
switch registers over MDIO, so on a board where those rails really drop
the teardown is talking to a switch that is already off. A devres disable
runs after .remove returns and would fix that too.

So I would keep this one as the NULL check with a Fixes tag, and send the
conversion to net-next, where the voltage handling and the power-up
ordering can be argued on their own.