Date: Mon, 3 Aug 2015 19:04:28 +0200
From: Oleg Nesterov <oleg@redhat.com>
To: Nikolay Borisov <kernel@kyup.com>
Cc: "Linux-Kernel@Vger. Kernel. Org" <linux-kernel@vger.kernel.org>,
        riel@redhat.com, Peter Zijlstra <peterz@infradead.org>,
        Steven Rostedt <rostedt@goodmis.org>
Subject: Re: HARD LOCKUP: Strange hard lock up on
	spin_lock(&sighand->siglock);
Message-ID: <20150803170428.GA18618@redhat.com>
References: <55B629E9.6020207@kyup.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <55B629E9.6020207@kyup.com>
User-Agent: Mutt/1.5.18 (2008-05-17)
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1242
Lines: 30

Sorry for delay, vacation.

I'll try to re-read your email later, just one note for now...

On 07/27, Nikolay Borisov wrote:
>
> Based on that I think could be happening is that the sighand itself is
> being freed while we are in the grace period inside __lock_task_sighand
> but the slab page itself is not freed as per the semantics of
> SLAB_DESTROY_BY_RCU. I looked up the source of this function in the
> latest kernels and saw that Oleg had put a comment clarifying the
> semantics but I'm still not convinced that it is safe. What if
> we are trying to lock the spinlock before this particular slab is
> initialised with sighand_ctor?

But this is not possible? ->sighand can never point to the uninitialized
struct sighand_struct.

Just in case... please note that if ->sighand was freed and then
re-allocated while __lock_task_sighand() spins under rcu_read_lock(),
sighand_ctor() won't be called again (due to SLAB_DESTROY_BY_RCU).
Perhaps this was the source of your confusion?

Oleg.

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/