Re: [PATCH 0/5] Rust abstractions for network device drivers

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

On Fri, Jun 16, 2023 at 8:26 PM Jakub Kicinski <kuba@kernel.org> wrote:
>
> Duplicated driver in a new language means nobody has a real incentive
> to use it in production. That really mutes the signal we get out of the
> experiment.

What I was trying to say is that there may be other incentives for
using the new one, like the ones I gave in the list.

Even if there is no incentive for using it right now, one may still
want to build/design it in order to evaluate Rust and/or prepare
abstractions for future drivers. This was the reason for the NVMe
driver request we got back then, for instance.

> At the same time IIUC building the Rust code is not trivial,
> so IDK if we're ready to force people to use it. Ugh.

I am not sure what you mean -- building works the same way as you have
done now, i.e. your usual build commands with `make`.

If you mean installing the toolchain, then we have a Quick Start guide
[1], and a script that checks for the requirements.

[1] https://docs.kernel.org/rust/quick-start.html

> Do you have any idea how long it will take until one can
>  dnf install $rust
> and have that be enough to be build a kernel (for the two major arches)?

You may do it today if the version matches, which for Fedora it should
e.g. next week or so when we merge the upgrade to 1.70, since Fedora
has now 1.70 [2].

But since they may not always align, I would recommend using `rustup`
(or the standalone installers) until we can establish a minimum
version. Please see the Quick Start guide for details [1].

Other distributions may provide a custom package with the required
dependencies for building Rust kernel code, like Ubuntu does.

[2] https://packages.fedoraproject.org/pkgs/rust/rust/

Cheers,
Miguel