Re: [PATCH] modules as shared objects

[Rusty Russell]

In message <[e-mail]> you write:
>   * Freeing the .init.* sections.  Also requires the link script.
>     I believe the proper solution is to use a different program
>     header with a different load tag:
> 
> 	#define PT_INIT_LOAD	0x60000001
> 
>     In this way we don't make a mistake and discard a segment that
>     we're not supposed to.

Were you thinking of dropping the dynamic at the same time
(ie. putting the dynamic and init sections in the tail using the
linker script)?  Hmm, I guess not since you 

I think an "mod_arch_free_partial(struct module *mod, void *start,
void *end)" function which archs could use to do whatever fancy magic
they wanted (eg. trim exception and really drop the pages, or your
vmalloc_drop_pages() function, or nothing at all).  x86_64, sparc,
sparc64, ppc and arm do/will have their own allocators already.

> +module_core_size(const ElfW(Ehdr) *hdr, struct module *module,

I *really* hate this ElfW(type) notation.  I know it's used in
binutils etc, but it's a visual bitch to parse.  Please please please
at least use the #defines or typedefs within the generic
kernel/module.c so it's readable.

Other than some gratuitous reformatting 8), looks really good.
Implementing PPC now.

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