@alswl: Did you know the Linux kernel has a feature introduced in 1997 that allows Apple Silicon Macs to efficiently run x86 containers? It's called binfmt_misc, introduced in Linux 2.1.43, nearly 30 years ago. The principle is extremely simple: Linux...

Did you know that the Linux kernel has a feature introduced in 1997 that enables Apple Silicon Macs to efficiently run x86 containers?

It’s called binfmt_misc, introduced in Linux 2.1.43 — nearly 30 years ago.

The principle is extremely simple: When the Linux kernel executes a binary, it first reads the magic number from the file header. If you have registered a rule in /proc/sys/fs/binfmt_misc/ — “for files with this magic, hand them over to a specific interpreter” — the kernel automatically delegates execution to that interpreter. It’s essentially the same mechanism as #!/bin/bash (shebang), but it matches binary file headers instead.

Originally, it was used to let Linux directly “run” Java class files and Windows PE executables. Later, QEMU used this mechanism for user-mode cross-architecture emulation — on ARM Linux, executing an x86 ELF directly, with the kernel automatically invoking qemu-x86_64 to translate.

Then Apple came along. When you launch a Linux VM using Colima / Lima on an Apple Silicon Mac (using macOS’s Virtualization.framework) with rosetta: true enabled, the following happens:

1. macOS mounts the Rosetta translator binary into the Linux VM’s /mnt/lima-rosetta/rosetta via virtio-fs
2. Inside the VM, a standard binfmt_misc rule is registered: when encountering x86_64 ELF magic (7f454c46...02003e00), hand it over to /mnt/lima-rosetta/rosetta
3. From then on, any execution of x86_64 binaries within the VM is automatically handed off to Rosetta for translation by the kernel

This is completely transparent to containers — docker run --platform linux/amd64 nginx: Docker pulls down an x86 image, the processes inside the container are x86 ELF binaries, and the Linux kernel automatically invokes Rosetta via binfmt_misc to translate and execute them. The container itself has no idea what’s happening.

Performance comparison:

• Rosetta approach: approximately 70-90% of native performance (JIT translation, instruction-level optimization)
• QEMU full virtualization approach: only about 10-30% (complete x86 CPU emulation)

A kernel feature designed in 1997 to run Java class files became, in 2024, the critical infrastructure for running x86 containers on Apple Silicon. Sometimes the most enduring architectural decisions are those that are simple and generic enough abstractions.

$ colima ssh
$ cat /proc/sys/fs/binfmt_misc/rosetta
enabled
interpreter /mnt/lima-rosetta/rosetta
flags: OCF
magic 7f454c4602010100000000000000000002003e00