Date: Sat, 17 Dec 2016 14:49:56 +0100
From: Peter Zijlstra <peterz@infradead.org>
To: Nicolai =?iso-8859-1?Q?H=E4hnle?= <nhaehnle@gmail.com>
Cc: linux-kernel@vger.kernel.org,
        Nicolai =?iso-8859-1?Q?H=E4hnle?= <Nicolai.Haehnle@amd.com>,
        Ingo Molnar <mingo@redhat.com>, Maarten Lankhorst <dev@mblankhorst.nl>,
        Daniel Vetter <daniel@ffwll.ch>,
        Chris Wilson <chris@chris-wilson.co.uk>,
        dri-devel@lists.freedesktop.org
Subject: Re: [PATCH v2 04/11] locking/ww_mutex: Set use_ww_ctx even when
 locking without a context
Message-ID: <20161217134956.GX3107@twins.programming.kicks-ass.net>
References: <1480601214-26583-1-git-send-email-nhaehnle@gmail.com>
 <1480601214-26583-5-git-send-email-nhaehnle@gmail.com>
 <20161206152537.GV3045@worktop.programming.kicks-ass.net>
 <239fa361-331a-a7b6-9a0d-a6baa19a5003@gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <239fa361-331a-a7b6-9a0d-a6baa19a5003@gmail.com>
User-Agent: Mutt/1.5.23.1 (2014-03-12)
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1877
Lines: 46

On Fri, Dec 16, 2016 at 02:17:25PM +0100, Nicolai H?hnle wrote:
> On 06.12.2016 16:25, Peter Zijlstra wrote:
> >On Thu, Dec 01, 2016 at 03:06:47PM +0100, Nicolai H?hnle wrote:
> >
> >>@@ -640,10 +640,11 @@ __mutex_lock_common(struct mutex *lock, long state, unsigned int subclass,
> >> 	struct mutex_waiter waiter;
> >> 	unsigned long flags;
> >> 	bool first = false;
> >>-	struct ww_mutex *ww;
> >> 	int ret;
> >>
> >>-	if (use_ww_ctx) {
> >>+	if (use_ww_ctx && ww_ctx) {
> >>+		struct ww_mutex *ww;
> >>+
> >> 		ww = container_of(lock, struct ww_mutex, base);
> >> 		if (unlikely(ww_ctx == READ_ONCE(ww->ctx)))
> >> 			return -EALREADY;
> >
> >So I don't see the point of removing *ww from the function scope, we can
> >still compute that container_of() even if !ww_ctx, right? That would
> >safe a ton of churn below, adding all those struct ww_mutex declarations
> >and container_of() casts.
> >
> >(and note that the container_of() is a fancy NO-OP because base is the
> >first member).
> 
> Sorry for taking so long to get back to you.
> 
> In my experience, the undefined behavior sanitizer in GCC for userspace
> programs complains about merely casting a pointer to the wrong type. I never
> went into the standards rabbit hole to figure out the details. It might be a
> C++ only thing (ubsan cannot tell the difference otherwise anyway), but that
> was the reason for doing the change in this more complicated way.

Note that C only has what C++ calls reinterpret_cast<>(). It cannot
complain about a 'wrong' cast, there is no such thing.

Also, container_of() works, irrespective of what C language says about
it -- note that the kernel in general hard relies on a lot of things C
calls undefined behaviour.

> Are you sure that this is defined behavior in C? If so, I'd be happy to go
> with the version that has less churn.

It should very much work with kernel C.