Re: [PATCH 2/2] module: fix bne2 "gave up waiting for init of module libcrc32c"

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

On Wed, 2 Jun 2010 02:05:10 pm Linus Torvalds wrote:
> 
> On Tue, 1 Jun 2010, Brandon Philips wrote:
> > 
> > FWIW, I tried this same idea initially and it breaks because the
> > kobject EEXIST is coming from mod_sysfs_init() which happens further
> > up in load_module() before the list_add_rcu().
> 
> Ahh, right. So we'd need to mvoe that down too. As Rusty says, the kobject 
> EEXIST warning should be fairly harmless - albeit annoying. Do things 
> actually _work_ with that thing apart from the kobject warning?
> 
> > I also tried the obvious variation of moving the list_add_rcu() up
> > to where the find_module is but got:
> 
> Yeah, I don't think moving it up will work, because then the module symbol 
> resolver will see itself before having set up the symbols. So I think we 
> need to expose it on the modules list only after having done the 
> "simplify_symbols()" etc.
> 
> I dunno.
> 
> Does this work?
> 
> (Caveat emptor - I tried to make sure the error cases nest correctly, and 
> it all compiles, but it's untested. As usual. Rusty mentioned a "see 
> testing patch", but I didn't see it. Maybe he did the same thing)

Yep, oops.  I am testing a similar thing, but a bit more of a cleanup
on the way.

Moved all the sysfs-exposing stuff to the end just after we put in the
list (and thus to after the find check).  In turn spawns a sysfs cleanup
(those funcs exposed for no good reason).  Also, your two-way deps patch
really wins since we need to walk the list to create the links.

Code soon...
Rusty