Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Tue, 6 Jun 2023 16:51:40 +1000
From:   Dave Chinner <david@fromorbit.com>
To:     Roman Gushchin <roman.gushchin@linux.dev>
Cc:     Kirill Tkhai <tkhai@ya.ru>, akpm@linux-foundation.org,
        vbabka@suse.cz, viro@zeniv.linux.org.uk, brauner@kernel.org,
        djwong@kernel.org, hughd@google.com, paulmck@kernel.org,
        muchun.song@linux.dev, linux-mm@kvack.org,
        linux-fsdevel@vger.kernel.org, linux-xfs@vger.kernel.org,
        linux-kernel@vger.kernel.org, zhengqi.arch@bytedance.com
Subject: Re: [PATCH v2 3/3] fs: Use delayed shrinker unregistration
Message-ID: <ZH7XfD/pBcWzhHcc@dread.disaster.area>
References: <168599103578.70911.9402374667983518835.stgit@pro.pro>
 <168599180526.70911.14606767590861123431.stgit@pro.pro>
 <ZH6AA72wOd4HKTKE@P9FQF9L96D>
 <ZH6K0McWBeCjaf16@dread.disaster.area>
 <ZH6ge3yiGAotYRR9@P9FQF9L96D>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <ZH6ge3yiGAotYRR9@P9FQF9L96D>
Precedence: bulk

On Mon, Jun 05, 2023 at 07:56:59PM -0700, Roman Gushchin wrote:
> On Tue, Jun 06, 2023 at 11:24:32AM +1000, Dave Chinner wrote:
> > On Mon, Jun 05, 2023 at 05:38:27PM -0700, Roman Gushchin wrote:
> > > Isn't it possible to hide it from a user and call the second part from a work
> > > context automatically?
> > 
> > Nope, because it has to be done before the struct shrinker is freed.
> > Those are embedded into other structures rather than being
> > dynamically allocated objects.
> 
> This part we might consider to revisit, if it helps to solve other problems.
> Having an extra memory allocation (or two) per mount-point doesn't look
> that expensive. Again, iff it helps with more important problems.

Ah, I guess if you're concerned about memory allocation overhead
during register_shrinker() calls then you really aren't familiar
with what register_shrinker() does on memcg and numa aware
shrinkers? 

Let's ignore the fact that we could roll the shrinker structure
allocation into the existing shrinker->nr_deferred array allocation
(so it's effectively a zero cost modification), and just look at
what a memcg enabled shrinker must initialise if it expands the
shrinker info array because the index returned from idr_alloc()
is larger than the current array:

	for each memcg {
		for_each_node {
			info = kvmalloc_node();
			rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_info, info);
		}
	}

Hmmmm?

So, there really isn't any additional cost, it completely decouples
the shrinker infrastructure from the subsystem shrinker
implementations, it enables the shrinker to control infrastructure
teardown independently of the subsystem that registered the
shrinker, and it still gives guarantees that the shrinker is never
run after unregister_shrinker() completes. What's not to like?

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com