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;
Content-Type: text/plain; charset="utf-8"
MIME-Version: 1.0
Content-Transfer-Encoding: 8BIT
To:     Thomas Gleixner <tglx@linutronix.de>
From:   Chris Wilson <chris@chris-wilson.co.uk>
In-Reply-To: <alpine.DEB.2.21.1902281105520.1821@nanos.tec.linutronix.de>
Cc:     Sebastian Andrzej Siewior <bigeasy@linutronix.de>,
        LKML <linux-kernel@vger.kernel.org>,
        David Airlie <airlied@linux.ie>,
        intel-gfx@lists.freedesktop.org, dri-devel@lists.freedesktop.org,
        Peter Zijlstra <peterz@infradead.org>
References: <20190212162857.20003-1-bigeasy@linutronix.de>
 <20190226160038.5vwk465gwzfwn75i@linutronix.de>
 <155134800668.5847.13013477917200660218@skylake-alporthouse-com>
 <alpine.DEB.2.21.1902281105520.1821@nanos.tec.linutronix.de>
Message-ID: <155134918155.5847.17130320586410453868@skylake-alporthouse-com>
User-Agent: alot/0.6
Subject: Re: [Intel-gfx] [PATCH] drm/i915/fence: Do not use TIMER_IRQSAFE
Date:   Thu, 28 Feb 2019 10:19:41 +0000
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Quoting Thomas Gleixner (2019-02-28 10:09:26)
> On Thu, 28 Feb 2019, Chris Wilson wrote:
> 
> > Quoting Sebastian Andrzej Siewior (2019-02-26 16:00:38)
> > > On 2019-02-12 17:28:57 [+0100], To linux-kernel@vger.kernel.org wrote:
> > > > The timer is initialized with TIMER_IRQSAFE flag. It does look like the
> > > > timer callback requires this flag at all. Its sole purpose is to ensure
> > > > synchronisation in the workqueue code.
> > > > 
> > > > Remove TIMER_IRQSAFE flag because it is not required.
> > > 
> > > ping
> > 
> > We call del_timer_sync from irq context, which mandates using
> > TIMER_IRQSAFE. Failure to do so results in lots of explosions across CI
> > -- every machine managed to hit the warning.
> > 
> > It may not be the best of api, but it's the only one available for the
> > driver to use...
> 
> The comment in the header files says clearly:
> 
>  * Note: The irq disabled callback execution is a special case for
>  * workqueue locking issues. It's not meant for executing random crap
>  * with interrupts disabled. Abuse is monitored!                                   
> 
> So what's so special in drm that you need to call del_timer_sync() from
> interrupt context?

There's no protection against fence signaling from inside interrupt
context, and a lot of pressure to do so.
-Chris