Re: [PATCH] Initcall depends

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

In message <Pine.LNX.4.44.0206171658470.22308-100000@chaos.physics.uiowa.edu> y
ou write:
> Parts of a solution could be (based on yours above).
> 
> o Define KBUILD_OBJECT during the build, which contains the
>   name of the module the file we're building will/would end up.
> 
> o Use that to rename the __initcall_whatever to
>   __initcall__module__whatever.
> 
> o Make a symlink tree pointing to the objects that will be linked in
>   (Basically $(obj-y))
> 
> o Go through the symlink tree and for all objects which export objects
>   and have __initcalls, record that relation.

Hmm... how about "put export symbols in __ksymtab in object file even
if CONFIG_MODULE=n", just discard them in the final link.

Then build up table from that (we're looking at the objects anyway).

> o For all __initcall__moduleA__ in the objects to be linked into 
>   vmlinux, find the object that defines it in the symlink tree (its name
>   will be moduleA.o).
> 
> o Find the unresolved symbols in that object moduleA.o.
> 
> o For each unresolved symbol in moduleA.o, if you find the symbol in the 
>   previously recorded pairs of (exported symbols, __initcall__moduleB), 
>   move __initcall_moduleA behind __initcall_moduleB.
> 
> Probably some issues come up when actually trying to do this, but it 
> sounds doable at least in principle.

I'll have to think some more and see if I can come up with something
along these lines.

> 
> Then again, there's also the possibility of completing initramsfs, making
> it mandatory, compiling things always as modules and leaving it to
> "depmod" in initramfs to do the right thing.

If that ever happens... 8)
Rusty.
--
  Anyone who quotes me in their sig is an idiot. -- Rusty Russell.