Re: [RFC][PATCH 0/5] do not dereference NULL pools in pools' destroy() functions

From: Andrew Morton <akpm@linux-foundation.org>
To: Christoph Lameter <cl@linux.com>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>,
	Minchan Kim <minchan@kernel.org>,
	Pekka Enberg <penberg@kernel.org>,
	Joonsoo Kim <iamjoonsoo.kim@lge.com>,
	Michal Hocko <mhocko@suse.cz>,
	David Rientjes <rientjes@google.com>,
	linux-mm@kvack.org, linux-kernel@vger.kernel.org,
	sergey.senozhatsky.work@gmail.com, Joe Perches <joe@perches.com>
Subject: Re: [RFC][PATCH 0/5] do not dereference NULL pools in pools' destroy() functions
Date: Tue, 9 Jun 2015 19:17:55 -0700	[thread overview]
Message-ID: <20150609191755.867a36c3.akpm@linux-foundation.org> (raw)
In-Reply-To: <alpine.DEB.2.11.1506092056570.6964@east.gentwo.org>

On Tue, 9 Jun 2015 21:00:58 -0500 (CDT) Christoph Lameter <cl@linux.com> wrote:

> On Tue, 9 Jun 2015, Andrew Morton wrote:
> 
> > > Why do this at all?
> >
> > For the third time: because there are approx 200 callsites which are
> > already doing it.
> 
> Did some grepping and I did see some call sites that do this but the
> majority has to do other processing as well.
> 
> 200 call sites? Do we have that many uses of caches? Typical prod system
> have ~190 caches active and the merging brings that down to half of that.

I didn't try terribly hard.

z:/usr/src/linux-4.1-rc7> grep -r -C1 kmem_cache_destroy .  | grep "if [(]" | wc -l
158

It's a lot, anyway.

> > More than half of the kmem_cache_destroy() callsites are declining that
> > value by open-coding the NULL test.  That's reality and we should recognize
> > it.
> 
> Well that may just indicate that we need to have a look at those
> callsites and the reason there to use a special cache at all.

This makes no sense.  Go look at the code. 
drivers/staging/lustre/lustre/llite/super25.c, for example.  It's all
in the basic unwind/recover/exit code.

> If the cache
> is just something that kmalloc can provide then why create a special
> cache. On the other hand if something special needs to be accomplished
> then it would make sense to have special processing on kmem_cache_destroy.

This has nothing to do with anything.  We're talking about a basic "if
I created this cache then destroy it" operation.

It's a common pattern.  mm/ exists to serve client code and as a lot of
client code is doing this, we should move it into mm/ so as to serve
client code better.