wiki/articles/microkernel.md

Microkernel

A microkernel is a kernel that has only the bare minimum functionality on kernel space, moving most of it to the userland, as servers. Due to having less code running in privileged kernel mode (less attack surface), microkernels can be more safer than monolithic ones.

A central concept of microkernels are the servers (daemons), which run in userland and implement the functionality that monolithic kernels do in the kernel side (file systems, drivers, memory management). Servers aren't different from regular user processes, only that have additional privileges (for accessing hardware directly, for example) and communicate with the kernel and with other servers using interprocess communication (IPC) mechanisms.

Since servers run in userspace, a crash in a server don't necessarily bring the whole system down, but may affect the consistency of the whole system, and reliability is not exactly inherent to the microkernel design.

One of the first microkernels, Mach, had a disappointing performance. This gave microkernels a bad reputation, which however where demonstrated to not be an inherent design characteristic, but an implementation one. The L4 family of microkernels demonstrated that IPC can be as fast as regular syscalls on a monolithic kernel.

Notable microkernels

• Mach
• L4 and its descendants, such as seL4; known for its high-performance IPC, as fast as a monolithic kernel.
• Minix: Unix-like kernel. Most notably, a heavily modified version of it runs in the Intel's ME backdoor.
• QNX: proprietary real time microkernel from the 80's.
• ...