Tracking: packages broken on larger pagesize kernels (like rpi5-kernel)

[classabbyamp]

Some programs have issues when running on kernels with larger pagesizes.

jemalloc

One common case is programs (especially rust programs) that use jemalloc. this was fixed in #48194 for the jemalloc package itself and all rust packages built in the future, but some packages still may need a rebuild. This often manifests itself like this:

<jemalloc>: Unsupported system page size
<jemalloc>: Unsupported system page size
memory allocation of 5 bytes failed

Probably broken

as yet untested, based on https://github.com/AsahiLinux/docs/wiki/Broken-Software#broken-packages

• [ ] qt5-webengine
• [ ] qt6-webengine (qt6-pdf?)

broken

• [ ] zig: https://github.com/ziglang/zig/pull/17382
• [ ] f2fs: broken in linux <=6.6, fixed in 6.7

[dkwo]

zig was definitely broken last year on 16k, the relevant pr's are in progress https://github.com/ziglang/zig/pull/17382

[dkwo]

For zig, we may consider this patch to mem.zig:

pub const page_size = switch (builtin.cpu.arch) {
    .wasm32, .wasm64 => 64 * 1024,
    .aarch64 => switch (builtin.os.tag) {
+        .macos, .ios, .watchos, .tvos, .linux => 16 * 1024,
-        .macos, .ios, .watchos, .tvos, => 16 * 1024,
        else => 4 * 1024,
    },
    .sparc64 => 8 * 1024,
    else => 4 * 1024,
};

I don't know how many aarch64 machines this would break, but at least it could be usable on rpi5 and apple.

[Calandracas606]

AFAIK F2FS still won't work, since it requires userspace support from f2fs-tools, which hasn't had a release yet.

[dkwo]

Surprisingly, river from void's repo works fine on apple silicon (16k kernel), even though zig is broken. How is this possible??

[Calandracas606]

Surprisingly, river from void's repo works fine on apple silicon (16k kernel), even though zig is broken. How is this possible??

this makes sense, and works for the same reason that any cross compiled binaries work.

the zig compiler may be broken when using a 16k build machine, but binaries built by the compiler may work just fine.

this is exactly the same as why cross compilers work. the aarch64 gcc cross compiler runs on a x86_64 build machine, and outputs binaries which run on an aarch64 machine