Re: system call stubs in glibc

[Andreas Jaeger <aj at suse dot de>]

Urs Thuermann <urs@isnogud.escape.de> writes:

> There are system calls which exist in the Linux kernel on some
> architectures but not others.  For example time(2) and stime(2) exist
> on all architectures but alpha and ia64.  I assume these are
> implemented by glibc using gettimeofday(2) and settimeofday(2) on
> alpha and ia64, correct?
>
> If so, how does glibc know, which system calls on which architectures
> exist, and when to implement them using other calls?  Is this
> configured in some architecture-depend config file or does glibc check
> itself for the availability of system calls like time(2) and stime(2)?

glibc reads <asm/unistd.h> and generates syscalls from this.  This is
a pretty complicated process.  To understand what's going check how
time is build now on your system.

> The reason I ask is, I'd like to know, if I would remove time() and
> stime() from the kernel on my i386/Linux, would glibc, when configured
> and compiled again running on that modified kernel, use
> get/settimeofday() or would it fail when someone uses time() or
> stime()?

What you need to do is:
- remove them from asm/unistd.h
- build a new glibc
- install the new glibc and the new kernel

i386/Linux might use sysdeps/unix/sysv/i386/time.S, check that instead
sysdeps/unix/time.c is used for compilation of time.os.

Note that the old glibc will not work with the new kernel.

Andreas
-- 
 Andreas Jaeger
  SuSE Labs aj@suse.de
   private aj@arthur.inka.de
    http://www.suse.de/~aj