kernel - Why is the setup syscall not callable by any user? -

http://linux.die.net/man/2/setup

i didn't find reason on man page.

first off, syscall no longer exists:

since linux 2.1.121, no such function exists anymore.

from can see in 2.1.120, used in init of kernel (and in fact has syscall number 0). called twice: first call freed initial memory (from initram) , second call sets devices, filesystems , mounted root. bad if program run syscall (it cause segmentation violations or other bad things).

why it's syscall in first place question. downloaded 1.0.0, , even then syscall not used init (which first guess why syscall). in 0.01 used inside kernel. since there's no usage int 0x80, i'm perplexed why linus did this.

i've sent email linus & lkml , asked on irc (although it's unlikely other linus know, , he's busy answering emails mitochondrial dna of linux).

edit: okay, looking further, turns out kernel calling process context (the setup call acts syscall, int 0x80 , all). initialisation of kernel not done within process context, , (according linus):

so code goes user space, initial user space shared kernel (until first fork()). initial user mode transition switching user segments [...] in init/main.c, magic that

move_to_user_mode(); if (!fork()) {          /* count on going ok */         init(); } for(;;) pause();        /* [this idle task] */ 

where "move_to_user_mode()" reload segments (some hand, cs/ss doing "iret"). first fork() done in user space, , before happens kernel cannot sleep (because there no idle task). [...] "setup()" system call because needs sleep (to io), , kernel couldn't sleep before got user-mode , first fork thing.

could have been done differently? sure. don't way more, , create idle tasks separately , not "fork()" more. kind of made sense @ time.