Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Fri, 22 Mar 2019 17:47:18 +0000
From:   Christopher Lameter <cl@linux.com>
To:     Waiman Long <longman@redhat.com>
cc:     Andrew Morton <akpm@linux-foundation.org>,
        Pekka Enberg <penberg@kernel.org>,
        David Rientjes <rientjes@google.com>,
        Joonsoo Kim <iamjoonsoo.kim@lge.com>,
        linux-kernel@vger.kernel.org, linux-mm@kvack.org,
        selinux@vger.kernel.org, Paul Moore <paul@paul-moore.com>,
        Stephen Smalley <sds@tycho.nsa.gov>,
        Eric Paris <eparis@parisplace.org>,
        "Peter Zijlstra (Intel)" <peterz@infradead.org>,
        Oleg Nesterov <oleg@redhat.com>
Subject: Re: [PATCH 1/4] mm: Implement kmem objects freeing queue
In-Reply-To: <20190321214512.11524-2-longman@redhat.com>
Message-ID: <01000169a683a0ed-3fa1b014-8efa-4c8f-a7e1-958e9eccd693-000000@email.amazonses.com>
References: <20190321214512.11524-1-longman@redhat.com> <20190321214512.11524-2-longman@redhat.com>
User-Agent: Alpine 2.21 (DEB 202 2017-01-01)
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Feedback-ID: 1.us-east-1.fQZZZ0Xtj2+TD7V5apTT/NrT6QKuPgzCT/IC7XYgDKI=:AmazonSES
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Thu, 21 Mar 2019, Waiman Long wrote:

> When releasing kernel data structures, freeing up the memory
> occupied by those objects is usually the last step. To avoid races,
> the release operation is commonly done with a lock held. However, the
> freeing operations do not need to be under lock, but are in many cases.
>
> In some complex cases where the locks protect many different memory
> objects, that can be a problem especially if some memory debugging
> features like KASAN are enabled. In those cases, freeing memory objects
> under lock can greatly lengthen the lock hold time. This can even lead
> to soft/hard lockups in some extreme cases.
>
> To make it easer to defer freeing memory objects until after unlock,
> a kernel memory freeing queue mechanism is now added. It is modelled
> after the wake_q mechanism for waking up tasks without holding a lock.

It is already pretty easy. You just store the pointer to the slab object
in a local variable, finish all the unlocks and then free the objects.
This is done in numerous places of the kernel.

I fear that the automated mechanism will make the code more difficult to
read and result in a loss of clarity of the sequencing of events in
releasing locks and objects.

Also there is already kfree_rcu which does a similar thing to what you are
proposing here and is used in numerous places.