From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <linux-kernel-owner@vger.kernel.org>
Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand
        id S1751761AbdIENgR (ORCPT <rfc822;w@1wt.eu>);
        Tue, 5 Sep 2017 09:36:17 -0400
Received: from Galois.linutronix.de ([146.0.238.70]:45583 "EHLO
        Galois.linutronix.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org
        with ESMTP id S1750932AbdIENgP (ORCPT
        <rfc822;linux-kernel@vger.kernel.org>);
        Tue, 5 Sep 2017 09:36:15 -0400
Date: Tue, 5 Sep 2017 15:36:11 +0200 (CEST)
From: Thomas Gleixner <tglx@linutronix.de>
To: Peter Zijlstra <peterz@infradead.org>
cc: mingo@kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH 0/2] smp/hotplug annotations
In-Reply-To: <alpine.DEB.2.20.1709051034000.1900@nanos>
Message-ID: <alpine.DEB.2.20.1709051534010.1900@nanos>
References: <20170905075218.526515965@infradead.org> <alpine.DEB.2.20.1709051034000.1900@nanos>
User-Agent: Alpine 2.20 (DEB 67 2015-01-07)
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Sender: linux-kernel-owner@vger.kernel.org
List-ID: <linux-kernel.vger.kernel.org>
X-Mailing-List: linux-kernel@vger.kernel.org

On Tue, 5 Sep 2017, Thomas Gleixner wrote:
> On Tue, 5 Sep 2017, Peter Zijlstra wrote:
> 
> > These two patches appear to make hotplug work again without tripping lockdep.
> 
> They cover the case where the plug/unplug succeeds, but they will not work
> when a plug/unplug operation fails, because after a fail it rolls back
> automatically, so in case UP fails, it will go down again, but the
> initiator side still waits on the 'UP' completion. Same issue on down.
> 
> I think that extra lockdep magic can be avoided completely by splitting the
> completions into a 'up' and a 'down' completion, but that only solves a
> part of the problem. The current failure handling does an automated
> rollback, so if UP fails somewhere the AP rolls back, which means it
> invokes the down callbacks. DOWN the other way round.
> 
> We can solve that by changing the way how rollback is handled so it does
> not automatically roll back.
> 
>     if (callback() < 0) {
>        store_state();
>        complete(UP);
>        wait_for_being_kicked_again()
>     }
> 
> and on the control side have
> 
>     wait_for_completion(UP);
> 
>     if (UP->failed) {
>        	kick(DOWN);
> 	wait_for_completion(DOWN);
>     }
> 
> It's not entirely trivial, but I haven't seen a real problem with it yet.

Now I found one. It's the multi instance rollback. This is a nested
rollback mechanism deep in the call chain. Seperating that one is going to
be a major pain.

Thanks,

	tglx