From mboxrd@z Thu Jan 1 00:00:00 1970 Date: Thu, 08 Nov 2007 15:24:08 -0800 (PST) Message-Id: <20071108.152408.157342087.davem@davemloft.net> Subject: Re: Some interesting observations when trying to optimize vmstat handling From: David Miller In-Reply-To: References: Mime-Version: 1.0 Content-Type: Text/Plain; charset=us-ascii Content-Transfer-Encoding: 7bit Sender: owner-linux-mm@kvack.org From: Christoph Lameter Date: Thu, 8 Nov 2007 11:58:58 -0800 (PST) Return-Path: To: clameter@sgi.com Cc: linux-kernel@vger.kernel.org, linux-mm@kvack.org, ak@suse.de, mathieu.desnoyers@polymtl.ca List-ID: > The problem with cmpxchg_local here is that the differential has to > be read before we execute the cmpxchg_local. So the cacheline is > acquired first in read mode and then made exclusive on executing the > cmpxchg_local. I bet this can be defeated by prefetching for a write before the read, but of course this won't help if the read is being used to conditionally avoid the cmpxchg_local but I don't think that's what you're trying to do here. I've always wanted to add a write prefetch at the beginning of all of the sparc64 atomic operation primitives because of this problem. I just never got around to measuring if it's worthwhile or not. -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@kvack.org. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: email@kvack.org