Re: [PATCH] Fix undefined/miscompiled construct in kernel parameters

[Richard Henderson <rth@twiddle.net>]

On Sun, Jun 15, 2003 at 07:48:56PM +0200, Roman Zippel wrote:
> Does the patch below work better?
> 
> bye, Roman
> 
> --- linux/init/main.c	14 Jun 2003 23:01:48 -0000	1.1.1.41
> +++ linux/init/main.c	15 Jun 2003 17:46:16 -0000
> @@ -383,7 +383,7 @@ asmlinkage void __init start_kernel(void
>  {
>  	char * command_line;
>  	extern char saved_command_line[];
> -	extern struct kernel_param __start___param, __stop___param;
> +	extern struct kernel_param __start___param[], __stop___param[];
>  /*
>   * Interrupts are still disabled. Do necessary setups, then
>   * enable them
> @@ -403,8 +403,8 @@ asmlinkage void __init start_kernel(void
>  	build_all_zonelists();
>  	page_alloc_init();
>  	printk("Kernel command line: %s\n", saved_command_line);
> -	parse_args("Booting kernel", command_line, &__start___param,
> -		   &__stop___param - &__start___param,
> +	parse_args("Booting kernel", command_line, __start___param,
> +		   __stop___param - __start___param,
>  		   &unknown_bootoption);
>  	trap_init();
>  	rcu_init();

Linus, I'd REALLY prefer this patch be applied, even though
the problem turned out to be one of amd64 alignment, which
can be worked around.

Even if struct kernel_param doesn't suffer from .sdata problems,
this formulation is closer to Correct.  I'd really prefer that
all such linker-script generated arrays used the [] form, and
not worry about the size of the data object involved.


r~