Date: Thu, 03 May 2012 03:02:42 -0400 (EDT)
Message-Id: <20120503.030242.2286535518319226860.davem@davemloft.net>
To: hpa@zytor.com
Cc: torvalds@linux-foundation.org, npiggin@gmail.com, jana@saout.de,
        jlbec@evilplan.org, linux-kernel@vger.kernel.org
Subject: Re: Oops with DCACHE_WORD_ACCESS and ocfs2, autofs4
From: David Miller <davem@davemloft.net>
In-Reply-To: <4FA22C75.6010001@zytor.com>
References: <4FA227FB.7060709@zytor.com>
	<20120503.025427.1277785208536786257.davem@davemloft.net>
	<4FA22C75.6010001@zytor.com>
Mime-Version: 1.0
Content-Type: Text/Plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1076
Lines: 27

From: "H. Peter Anvin" <hpa@zytor.com>
Date: Wed, 02 May 2012 23:57:57 -0700

> On 05/02/2012 11:54 PM, David Miller wrote:
>>>
>>> Can we do the trick of aligning the pointer and ignoring the start?
>>> That would allow even architectures that don't have unaligned accesses
>>> to work, too.
>> 
>> Doing that would flub the hash computation.
> 
> I guess the shifts would be to expensive?

Yes, barrel-shifting (if that's your idea) would negate much of the
gain from the optimization.

Actually, thinking some more, a barrel-shifting loop would have the
same problem the current code has.  You don't know if you are at the
end of the string until you do the tests on the word.  But if you're
at the end of the page, you need to somehow elide that extra load
to get the word you're going to barrel-shift into the previous word.

--
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/