Re: [PATCH] automatic initcalls

[Rusty Russell <rusty@rustcorp.com.au>]

In message <Pine.LNX.4.44.0207272145050.6125-100000@home.transmeta.com> you wri
te:
> 
> 
> On Sat, 27 Jul 2002, Jeff Garzik wrote:
> >
> > I've always preferred a system where one simply lists dependencies [as
> > you describe above], and some program actually does the hard work of
> > chasing down all the initcall dependency checking and ordering.
> >
> > Linus has traditionally poo-pooed this so I haven't put any work towards
> > it...
> 
> I don't hate the notion, but at the same time every time it comes up I
> feel that there are reasonably simple ways to just avoid the ordering
> problems.

I think that the best hope is a combination of Roman's module depends
work (based on Kai's "everything which is a module is trivial", and
Stephen and my first depends hack) and explicit depends.

Linkage ordering doesn't work in general, for things like "I want to
be initialized before the non-boot cpus have come up", but for
non-core code it's simple.

> Rusty had a script, but somebody complained about the speed of it. I
> haven't looked at it myself.

Yes, it'll slow the build by a few seconds: but if the linker ever
decides not to preserve ordering, we'll need it.  My original shell
script is suboptimal but we *don't* want the kernel build relying on
libbfd.

Roman and I will come up with something and send it to you later this
week.

Thanks!
Rusty.
--
  Anyone who quotes me in their sig is an idiot. -- Rusty Russell.