[okl4-developer] x86 syscall implementation

[Carlos Eduardo Medaglia Dyonisio]

On Fri, 31 Oct 2008 10:52:36 +0100
Harry Vennik <htvennik at zonnet.nl> wrote:

> Hi,
> 
> While trying to understand the syscall meganism for x86 I ran into  
> one remarkable line in the implementation of all syscalls in arch/ 
> ia32/libs/l4/include/syscall.h (okl4 version 2.1).
> 
> It is the following assembly line:
> 
>              "       movl $0f, %%edi\n"
> 
> what does the $0f mean??? Is it a simple hex value? (should be $0xf  
> then, i suppose) or is it something different?
> And why is that value loaded into the EDI register?
> 
> Regards,
> 
> Harry
>

Hello Harry,

Because of the way sysenter/sysexit works on ia-32, we have to pass the
user's return IP and SP to kernel space in registers. In our implementation
we chose to pass the user's return IP on %edi and the SP on %ebp. %eax
contains the syscall number.

Here is a piece of the code you are talking about:

        __SYSCALL_SAVE_REGS
"       movl $0f, %%edi\n"
"       movl %%esp, %%ebp\n"
"       movl $0x80000004, %%eax\n"
"       sysenter\n"
"0:\n"
        __SYSCALL_RESTORE_REGS

$0f is the address of the first tag 0, searching forward from the point where
0 was referenced (this is the reason of the 'f'. If it was for gcc to search
for the first tag 0 backwards, it would be a 'b'), which is the tag just
after sysenter, to where the syscall should return.

I hope I made it clear... :)

Cheers,
Carlos
-- 
Carlos Eduardo Medaglia Dyonisio
medaglia at ok-labs.com
http://ok-labs.com/