Date: Mon, 4 Aug 2008 23:54:02 +0200
From: Andrea Arcangeli <andrea@qumranet.com>
To: Arjan van de Ven <arjan@infradead.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>, Dave Jones <davej@redhat.com>,
       Roland Dreier <rdreier@cisco.com>,
       Linus Torvalds <torvalds@linux-foundation.org>,
       David Miller <davem@davemloft.net>, jeremy@goop.org, hugh@veritas.com,
       mingo@elte.hu, akpm@linux-foundation.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH] workaround minor lockdep bug triggered by
	mm_take_all_locks
Message-ID: <20080804215402.GE12464@duo.random>
References: <20080804162657.GI11476@duo.random> <1217867935.3589.35.camel@twins> <20080804172728.GJ11476@duo.random> <20080804174659.GK11476@duo.random> <20080804175730.GL11476@duo.random> <1217875739.3589.56.camel@twins> <20080804201514.GB12464@duo.random> <1217882242.3589.90.camel@twins> <20080804210954.GC12464@duo.random> <20080804142752.23a59073@infradead.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20080804142752.23a59073@infradead.org>
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1681
Lines: 34

On Mon, Aug 04, 2008 at 02:27:52PM -0700, Arjan van de Ven wrote:
> I think you totally misunderstand things then.
> 
> Lockdep will report a problem if it *ever* sees a BA order after it has
> seen a BA order. They don't have to happen at the same time. Or even
> within hours of eachother.

For me the false positive was generated by the recursive spinlock
feature, not the deadlock inversion feature. If you limit my comments
to the recursive spinlock feature perhaps you'll be more likely to
agree with them.

> They MIGHT happen... in a narrow time window, when you have a deadlock.
> But lockdep will warn you about the order violation without actually
> having to dealock... because the AB is likely to be done already most
> of the time when the BA happens.

So this is about the lock inversion feature, I admit I didn't realize
it has memory that locks have been taken in AB order and spawns the
first time locks have been taken in BA order. But this will lead to
more false positives because there's nothing wrong to do AB BA if
there's C lock taken before them!

Perhaps enforcing lockdep all over the kernel is worth it just for
this AB BA thing in case it doesn't only spawn false positives, but
still I can't like something that is as inaccurate and prone for
errors as lockdep and spreading all over the place to try to reduce
the false positives it emits. I can't like that but that's just
me... others clearly love it.
--
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/