From mboxrd@z Thu Jan 1 00:00:00 1970 Date: Mon, 12 Jun 2006 14:13:41 -0700 (PDT) From: Christoph Lameter Message-Id: <20060612211341.20862.95937.sendpatchset@schroedinger.engr.sgi.com> In-Reply-To: <20060612211244.20862.41106.sendpatchset@schroedinger.engr.sgi.com> References: <20060612211244.20862.41106.sendpatchset@schroedinger.engr.sgi.com> Subject: [PATCH 11/21] swap_prefetch: Conversion of nr_slab to ZVC Sender: owner-linux-mm@kvack.org Return-Path: To: linux-kernel@vger.kernel.org Cc: akpm@osdl.org, Hugh Dickins , Con Kolivas , Marcelo Tosatti , Nick Piggin , linux-mm@kvack.org, Andi Kleen , Dave Chinner , Christoph Lameter List-ID: This removes a potential problem for swap_prefetch. Use NR_SLAB and remove the comment stating the problem. Signed-off-by: Christoph Lameter Index: linux-2.6.17-rc6-cl/mm/swap_prefetch.c =================================================================== --- linux-2.6.17-rc6-cl.orig/mm/swap_prefetch.c 2006-06-12 11:30:59.995187077 -0700 +++ linux-2.6.17-rc6-cl/mm/swap_prefetch.c 2006-06-12 11:55:06.458392224 -0700 @@ -389,14 +389,11 @@ static int prefetch_suitable(void) /* * >2/3 of the ram on this node is mapped, slab, swapcache or * dirty, we need to leave some free for pagecache. - * Note that currently nr_slab is innacurate on numa because - * nr_slab is incremented on the node doing the accounting - * even if the slab is being allocated on a remote node. This - * would be expensive to fix and not of great significance. */ limit = node_page_state(node, NR_MAPPED) + node_page_state(node, NR_ANON) + - ps.nr_slab + ps.nr_dirty + + node_page_state(node, NR_SLAB) + + ps.nr_dirty + ps.nr_unstable + total_swapcache_pages; if (limit > ns->prefetch_watermark) { node_clear(node, sp_stat.prefetch_nodes); -- 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