How the ARM32 Kernel Starts | Hacker News Reader

How the ARM32 Kernel Starts

people.kernel.org

84 pointstrulyrandom9 comments

This article starts out talking about Linux address space layout. I've heard Solaris sometimes allocated userspace pages in the upper half of 64-bit address spaces. I'm guessing that by this point, it's too baked into Linux to change.

But, if you ever find yourself writing an OS, consider running userspace as much as possible near the top of a 64-bit address space. When the first 13 bits of a 64-bit pointer are set, a pointer interpreted as an IEEE-754 double comes out as NaN. The NaNboxing an NuNboxing tricks used by many dynamically typed programming languages become no-ops if all of your userspace pages are in the top 2 petabytes of the address space.

On the other hand, if you ever find yourself writing a programming language, consider making it statically typed.

In the world of mobile devices where CPU is battery life, why create a language that is ~10x slower for no user benefit and makes people contemplate about designing the whole OS around its performance issues?

Well, NaN boxing/NuN boxing would also allow you to more densely encode algebraic data type variants in statically typed languages in the case where at least one variant is a IEEE-754 double, some of the variants are references/pointers, and none of the variants take more than 8 bytes.

Granted, it's not a common use case. Neither Rust nor OCaml use this space optimization, but they could. OCaml (at least used to, back in the early 2000s) made the maximum tuple/object size smaller in other to have some bits (8 bits, IIRC) to store the variant in the object header. I'm not actually sure how Rust decides how many bits to dedicate to encoding the variant, and I have very little knowledge of GHC internals.

I'm a big fan of Rust (and to some extent, C++), but there's a ton of CPU time being spent on JavaScript execution worldwide.

So, the 'linusw' in the link[1] means it's an article of Linus, or some people else also named Linus?

[1] https://people.kernel.org/linusw/how-the-arm32-kernel-starts

https://github.com/linusw

You are right. Linus Walleij's name is on the list of kernel maintainers[1].

[1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/lin...

I believe that is Linus Walleij.

[Linux], no?

Yes, kernel.org is the archive for the Linux kernel. I didn't realize that they publish articles though, which is pretty neat.

Reply on news.ycombinator.com