Re: [PATCH] Many groups patch.

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

In message <20030929211155.A28089@hockin.org> you write:
> On Tue, Sep 30, 2003 at 09:30:07AM +1000, Rusty Russell wrote:
> > > Why?
> > 
> > (Rusty points at Tim).
> > 
> > He has 10,000 groups.  Now me, I'm happy with the minimal fix.
> 
> I'm going to merge your thoughts and mine tomorrow and send it out.  Linus
> suggested the array of pages approah is more sane, so I'm going to try for
> it.  I'm going to comb through the diffs between your patch and mine.

Keeping the groups array as an array is a feature.  IMHO, if there are
too many for a kmalloc, vmalloc fallback makes sense: it's
conceptually simple and not that much code.

Introducing an almost-vmalloc because Linus didn't like the vmalloc
just doesn't make sense, IMHO.

> > And worse, there are the intermediate kmallocs which would need to be
> > fixed (thanks to Stephen Rothwell for pointing this out).  Fixing this
> > would make it even uglier.
> 
> Specifically?  I think my patch gets all of those.  At least all the ones I
> found.

Yep, you got them, I didn't.  I am still hoping Stephen (CC'd) will
move the IA64, S390 and Sparc64 code into kernel/compat.c which will
shrink out patches.

> > Here's an updated one (with David Meybohm's fix, too -- Thanks!),
> > Rusty.
> 
> Can you elaborate on what this extra fix is?

I screwed up a free.

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