Re: [start_kernel] Suggest to move parse_args() before trap_init()

[Dave Hansen <haveblue@us.ibm.com>]

On Mon, 2004-03-01 at 12:46, Andi Kleen wrote:
> Andrew Morton <akpm@osdl.org> writes:
> 
> > I think the only problem with this is if we get a fault during
> > parse_args(), the kernel flies off into outer space.  So you lose some
> > debuggability when using an early console.
> > 
> > But 2.4 does trap_init() after parse_args() and nobody has complained, as
> > did 2.6 until recently.  So the change is probably OK.
>...
> Another way that i've considered on x86-64 for 2.7 at least is a special
> __early_setup() for this

It might be nice to have the same thing as the initcalls:
#define core_initcall(fn)               __define_initcall("1",fn)
...
#define late_initcall(fn)               __define_initcall("7",fn)

If we had something like this, we could also use it for stuff like
early_printk(), in addition to things like mem= in setup_arch().  You
can actually do serial early printk in about 3 lines of code if you can
parse arguments really, really early.  

-- dave