More like much later. There are several performance-critical patterns used in the Linux kernel that are difficult to express in unsafe Rust and impossible to express in safe Rust.
See https://paulmck.livejournal.com/62436.html and especially https://paulmck.livejournal.com/64209.html
I think promoting the use of Rust for driver code is an excellent compromise.
As an aside, I'd like to see Linux go further and move many of these drivers to user space. Now that we have io_uring, old performance-based arguments for keeping certain drivers in ring 0 hold less weight.
Linux's own support for the long tail of more obscure platforms is also an ongoing process. People aren't asking for Rust to do better than the Linux kernel in that regard, they're just asking for it to not immediately disqualify a range of platforms.
One could reasonably argue about the value of supporting those platforms, but the Linux kernel, reasonably, doesn't want to make that decision on the sole basis of the availability of Rust support.
> And for exceptionally weird platforms (e.g. where a byte isn't 8 bits) Rust may never bother to contort itself enough to support those, although perhaps Linux also doesn't support such platforms?
Linux doesn't support such platforms, and I don't think Rust ever will either.
Almost all software not written especially for such devices doesn't support non-8-platforms. Heck, POSIX mandates that char is 8 bits.
This is partially what caused the shitstorm when the python Cryptography library took a hard dependency on rust to compile:
You can still run 31-bit s390 usermode binaries on an s390x kernel, but that's a different beast.
Ah thanks, I didn't realize that was possible. I'm no where near rich and/or old enough to have ever encountered an s390 machine IRL :)
This means that using it for the core parts may be feasible sooner rather than later, no? Developing those abstractions may not be a priority of course, but it might be allowable at least from a platform support perspective.