setup(2)

Purpose

setup() was the one-shot bootstrap syscall: called once from linux/init/main.c, it ran the device and filesystem initialization functions configured into the kernel and mounted the root filesystem (setup.2, DESCRIPTION). No user process may call it – “Any user process, even a process with superuser permission, will receive EPERM” (setup.2, DESCRIPTION).

Synopsis

#include <unistd.h>
[[deprecated]] int setup(void);

Arguments and behavior

The man page records that the calling sequence varied over time – sometimes a single void *BIOS argument, sometimes an int magic – before removal in 2.1.121 (setup.2, VERSIONS). What modern kernels do at that point of boot (device init, rootfs mount) is in-kernel initcall machinery; the syscall slot survived as an i386 table fossil.

Return value and errors

Always -1 for a user process (setup.2, RETURN VALUE). Errors: EPERM, always, for user processes (setup.2, ERRORS).

Flags and constants

None.

Security and notes

Like idle, setup() is privileged-by-identity: the check is “are you process 0 during boot”, not a capability. Interview angle: these two slots are the answer to “did Linux ever have syscalls only the kernel’s own bootstrap could call?” – yes, and both were retired in the 2.1/2.3 era. Cross-reference: idle, unimplemented, reboot.

References

• https://man7.org/linux/man-pages/man2/setup.2.html
• man-pages syscalls(2) (setup listed among the removed/bootstrap calls)