Re: The time has come for a kernel interfacing library layer

[Matthew Dillon <dillon@xxxxxxxxxxxxxxxxxxxx>]

:>     Joerg, to make this work I need two other things:
:> 
:>     * We need to have the kernel automatically setup the initial TLS
:>       space.
:
:Why? Do you want to completely eliminate the syscall wrappers from libc?
:That's going to add problems later I fear. I'd instead change it the
:interface between library layer and syscall functions in libc to always
:provide a pointer to errno as argument. That would solve the problems
:e.g. with RTLD too.
:
:Joerg

    I considered providing a poiner to errno, but it would mean changing
    the int $0x80 algorithm which right now assumes that the arguments are
    a certain offset relative to the call stack.

    It's easier to simply require that a pointer to errno be stored in
    the TLS (via i386/include/tls.h).

    I've almost got this working.  I'm writing a new system call 
    configuration file format and a program to generate the fixed-offset
    shims.

    Basically as far as libc or rtld is concerned, they simply link
    against another library called 'libsys' (rtld gets its own, which we
    make the one libc uses part of the crt link sequence).  This library
    will contain the system calls in a specially-named section.  The
    kernel will then remap the section as appropriate.  We have the option
    of making the section a BSS section (so it takes up no space in the
    actual linked program), or making the section a code section with real
    'default' syscall entry points in it.

    Here's an example of a BSS code section reserving fixed-offsets for
    system calls:

        .section special,"ax", @nobits
        .globl  fubar0
        .globl  fubar1
        .globl  fubar2

        .p2align        12
	.org		0
syscall0:
	.org		64
syscall1:
	.org		128
syscall2:
        .org            192
        .p2align        12

    We generate a 'real code' version for the kernel to mmap-over the
    space the 'user bss' version reserves in the program for userland.
    Or, alternatively, we link the real code version into userland as
    a default and still allow the kernel to override it with its own
    'real code' version.

    I am gunning for the BSS version myself, because it takes no extra
    space in the program binary, but I haven't worked out all the 
    backwards compatibility issues yet.  It could very well be that we
    will have to have a third 'compact' syscall library just for rtld,
    sorta like we do now with the compact libc library we link against
    rtld.
    
					-Matt
					Matthew Dillon 
					<dillon@xxxxxxxxxxxxx>