Re: [PATCH v4 4/6] rust: irq: add support for threaded IRQs and handlers

[Miguel Ojeda <miguel.ojeda.sandonis@gmail.com>]

On Mon, Jun 9, 2025 at 8:13 PM Danilo Krummrich <dakr@kernel.org> wrote:
>
> On Mon, Jun 09, 2025 at 01:24:40PM -0300, Daniel Almeida wrote:
> >
> > This iteration converted register() from pub to pub(crate). The idea was to
> > force drivers to use the accessors. I assumed this was enough to make the API
> > safe, as the few users in the kernel crate (i.e.: so far platform and pci)
> > could be manually checked for correctness.
> >
> > To summarize my point, there is still the possibility of misusing this from the
> > kernel crate itself, but that is no longer possible from a driver's
> > perspective.
>
> Correct, you made Registration::new() crate private, such that drivers can't
> access it anymore. But that doesn't make the function safe by itself. It's still
> unsafe to be used from platform::Device and pci::Device.

Yeah.

Even if a function is fully private (i.e. not even `pub(crate)`), then
it should still be marked as unsafe if it is so.

Cheers,
Miguel