From mboxrd@z Thu Jan  1 00:00:00 1970
From: Eric Dumazet <dada1@cosmosbay.com>
Subject: Re: [PATCH 01 of 12] Core of mmu notifiers
Date: Tue, 22 Apr 2008 17:37:38 +0200
Message-ID: <480E0642.6080109@cosmosbay.com>
References: <ea87c15371b1bd49380c.1208872277@duo.random> <480DFC8A.8040105@cosmosbay.com> <20080422151529.GE24536@duo.random>
Mime-Version: 1.0
Content-Type: text/plain; charset=ISO-8859-1;
	format=flowed
Content-Transfer-Encoding: QUOTED-PRINTABLE
Cc: Christoph Lameter <clameter@sgi.com>,
	Nick Piggin <npiggin@suse.de>, Jack Steiner <steiner@sgi.com>,
	Peter Zijlstra <a.p.zijlstra@chello.nl>,
	kvm-devel@lists.sourceforge.net,
	Kanoj Sarcar <kanojsarcar@yahoo.com>,
	Roland Dreier <rdreier@cisco.com>,
	Steve Wise <swise@opengridcomputing.com>,
	linux-kernel@vger.kernel.org, Avi Kivity <avi@qumranet.com>,
	linux-mm@kvack.org, Robin Holt <holt@sgi.com>,
	general@lists.openfabrics.org, Hugh Dickins <hugh@veritas.com>,
	akpm@linux-foundation.org, Rusty Russell <rusty@rustcorp.com.au>
To: Andrea Arcangeli <andrea@qumranet.com>
Return-path: <linux-kernel-owner+glk-linux-kernel-3=40m.gmane.org-S1763220AbYDVPiB@vger.kernel.org>
In-Reply-To: <20080422151529.GE24536@duo.random>
Sender: linux-kernel-owner@vger.kernel.org
List-Id: kvm.vger.kernel.org

Andrea Arcangeli a =E9crit :
> On Tue, Apr 22, 2008 at 04:56:10PM +0200, Eric Dumazet wrote:
>  =20
>> Andrea Arcangeli a =E9crit :
>>    =20
>>> +
>>> +static int mm_lock_cmp(const void *a, const void *b)
>>> +{
>>> +	cond_resched();
>>> +	if ((unsigned long)*(spinlock_t **)a <
>>> +	    (unsigned long)*(spinlock_t **)b)
>>> +		return -1;
>>> +	else if (a =3D=3D b)
>>> +		return 0;
>>> +	else
>>> +		return 1;
>>> +}
>>> +
>>>      =20
>> This compare function looks unusual...
>> It should work, but sort() could be faster if the
>> if (a =3D=3D b) test had a chance to be true eventually...
>>    =20
>
> Hmm, are you saying my mm_lock_cmp won't return 0 if a=3D=3Db?
>  =20
I am saying your intent was probably to test

else if ((unsigned long)*(spinlock_t **)a =3D=3D
	    (unsigned long)*(spinlock_t **)b)
		return 0;


Because a and b are pointers to the data you want to compare. You need=20
to dereference them.


>> static int mm_lock_cmp(const void *a, const void *b)
>> {
>> 	unsigned long la =3D (unsigned long)*(spinlock_t **)a;
>> 	unsigned long lb =3D (unsigned long)*(spinlock_t **)b;
>>
>> 	cond_resched();
>> 	if (la < lb)
>> 		return -1;
>> 	if (la > lb)
>> 		return 1;
>> 	return 0;
>> }
>>    =20
>
> If your intent is to use the assumption that there are going to be fe=
w
> equal entries, you should have used likely(la > lb) to signal it's
> rarely going to return zero or gcc is likely free to do whatever it
> wants with the above. Overall that function is such a slow path that
> this is going to be lost in the noise. My suggestion would be to defe=
r
> microoptimizations like this after 1/12 will be applied to mainline.
>
> Thanks!
>
>  =20
Hum, it's not a micro-optimization, but a bug fix. :)

Sorry if it was not clear