Re: [PATCH] getpid/vfork/raise fix

[Jakub Jelinek <jakub at redhat dot com>]

On Wed, Mar 10, 2004 at 11:37:43AM +0100, Andreas Schwab wrote:
> --- nptl/sysdeps/unix/sysv/linux/ia64/pt-vfork.S.~1.2.~	2003-12-11 11:35:12.000000000 +0100
> +++ nptl/sysdeps/unix/sysv/linux/ia64/pt-vfork.S	2004-03-10 11:28:34.057587179 +0100
> @@ -1,4 +1,4 @@
> -/* Copyright (C) 2000, 2002, 2003 Free Software Foundation, Inc.
> +/* Copyright (C) 2000, 2002, 2003, 2004 Free Software Foundation, Inc.
>     This file is part of the GNU C Library.
>  
>     The GNU C Library is free software; you can redistribute it and/or
> @@ -32,13 +32,22 @@
>  ENTRY(__vfork)
>  	.prologue	// work around a GAS bug which triggers if
>  	.body		// first .prologue is not at the beginning of proc.
> -	alloc r2=ar.pfs,0,0,2,0
> +	alloc r2=ar.pfs,0,1,2,0
> +	adds r14=PID,r13
> +	;; 
> +	ld4 loc0=[r14]
> +	;;
> +	sub r15=0,loc0
>  	mov out0=CLONE_VM+CLONE_VFORK+SIGCHLD
>  	mov out1=0		/* Standard sp value.			*/
>  	;;
> +	st4 [r14]=r15
>  	DO_CALL (SYS_ify (clone))
> +	cmp.eq p0,p7=0,r8
>  	cmp.eq p6,p0=-1,r10
> +	adds r14=PID,r13
>  	;;
> +(p7)	st4 [r14]=loc0

Can you use a local register for saving oldval accross the syscall though?
Registers in the register window can be saved to the backing store, if that
happens, vfork child restores it from the backing store, calls some routine
and something else is stored into that location in backing store, vfork
parent will afterwards restore a different value.
E.g. IA-32 vfork saves return address in a register and not on the stack for
a reason.

IMHO only a call clobbered register preserved accross system call can be
used, and is it is not available, vfork has to choose a slower way:

int oldval = self->pid;
int newval = oldval ? -oldval : 0x80000000;
self->pid = newval;
vfork syscall
newval = self->pid;
oldval = (newval & 0x7fffffff) ? -newval : 0;
self->pid = oldval;

(and with pt-vfork just self->pid = -self->pid; before and after).

	Jakub