Hi Pekka,
On Tue, Mar 21, 2006 at 09:05:14AM +0200, Pekka J Enberg wrote:
> On Tue, 21 Mar 2006, Nathan Scott wrote:
> > Sorry, but thats just silly. Did you even look at the code
> > around what you're changing (it has to do more than just wrap
> > up slab calls)? So, NACK on this patch - it leaves the code
> > very confused (half zoney, half slaby), and is just unhelpful
> > code churn at the end of the day.
>
> You're already using kmem_cache_destroy() mixed with the zone stuff so I
> don't see your point.
Hmm, those ones have slipped through - I'll fix those up, thanks.
The vast majority of zone alloc/free uses are via the other style,
many variables are named according to those interfaces, etc, etc.
> I would really prefer to feed small bits at a time
> so is there any way I can sweet-talk you into merging the patch?
Not without a very convincing argument, and "just kill wrappers
without even looking at the existing code" hasn't made me very
enthusiastic so far.
Regarding __GFP_NOFAIL, there are some situations where that could
be used now, and others where it should not be - it'd take a very
careful code audit and evidence of a level of low-memory-condition
testing done, etc, before such a change would be merged. Note that
the -mm tree currently has a rework of the way incore extents are
managed within XFS, which significantly changes (in a good way) the
nature of the allocation requests we make (and hence I'm _really_,
_really_ not interested in cosmetic patches in this area just now).
cheers.
--
Nathan
|