Re: [PATCH net-next v2 9/9] dpll: document locking rules
From: Ivan Vecera
Date: Mon Nov 16 2026 - 03:23:50 EST
Sashiko reviewed v2 only on its dashboard [1], with low severity
findings, so I am answering them here.
[1] https://netdev-ai.bots.linux.dev/sashiko/#/patchset/20261111164021.273409-1-ivecera%40redhat.com
> Is this rule broad enough? Several other exported functions in
> drivers/dpll/dpll_core.c also take dpll_lock for write, and most of them
> do not change the topology:
[...]
> Would it make sense to say that these contexts must not call any exported
> dpll function that takes dpll_lock (everything except the __ prefixed
> notifiers), or to list these functions?
Yes. Every exported dpll function except the __ prefixed notifiers and
the netdev pin helpers takes dpll_lock, and rules 1 and 3 should cover
all of them. This is not new: on main, a callback calling
fwnode_dpll_pin_find() deadlocks on the dpll_lock mutex the same way. I
will widen both rules in a follow-up, or in v3 if one is needed.
> Should this section also describe the context of in-kernel notifier
> callbacks registered with register_dpll_notifier()? The rules only cover
> ops callbacks and driver code.
[...]
> Today ice_dpll_pin_notify() handles this correctly. It only allocates the
> work and queues it:
[...]
> The dpll calls are then made from ice_dpll_pin_notify_work(). Could this
> requirement be stated in the new rules as well?
Yes. Notifier callbacks run under dpll_lock, and with this series also
under the domain lock of the notified object, which can belong to
another driver. They have to follow the rules of a callback, and a
driver must not hold a lock they take while calling a function that
takes dpll_lock. I will add both to the same follow-up.
Thanks,
Ivan