Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1752033Ab1DOIVA (ORCPT ); Fri, 15 Apr 2011 04:21:00 -0400 Received: from cantor.suse.de ([195.135.220.2]:53830 "EHLO mx1.suse.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751901Ab1DOIU5 (ORCPT ); Fri, 15 Apr 2011 04:20:57 -0400 Date: Fri, 15 Apr 2011 10:20:51 +0200 From: Michal Hocko To: KOSAKI Motohiro Cc: LKML , Jack Steiner , Lee Schermerhorn , Christoph Lameter , Pekka Enberg , Paul Menage , Robin Holt , Andrew Morton , Linus Torvalds , linux-mm@kvack.org Subject: [PATCH v2] cpusets: randomize node rotor used in cpuset_mem_spread_node() Message-ID: <20110415082051.GB8828@tiehlicka.suse.cz> References: <20110414065146.GA19685@tiehlicka.suse.cz> <20110414160145.0830.A69D9226@jp.fujitsu.com> <20110415161831.12F8.A69D9226@jp.fujitsu.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20110415161831.12F8.A69D9226@jp.fujitsu.com> User-Agent: Mutt/1.5.21 (2010-09-15) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 7072 Lines: 219 [I just realized that I forgot to CC mm mailing list] On Fri 15-04-11 16:18:45, KOSAKI Motohiro wrote: > Oops. > I should have look into !mempolicy part too. > I'm sorry. > [...] > Michal, I think this should be > > #ifdef CONFIG_CPUSETS > if (cpuset_do_page_mem_spread()) > p->cpuset_mem_spread_rotor = node_random(&p->mems_allowed); > if (cpuset_do_slab_mem_spread()) > p->cpuset_slab_spread_rotor = node_random(&p->mems_allowed); > #endif > > because 99.999% people don't use cpuset's spread mem/slab feature and > get_random_int() isn't zero cost. > > What do you think? You are right. I was thinking about lazy approach and initialize those values when they are used for the first time. What about the patch below? Change from v1: - initialize cpuset_{mem,slab}_spread_rotor lazily --- >From 46da525789a41cb10096aa09c5bc0e7dcc3ce6db Mon Sep 17 00:00:00 2001 From: Michal Hocko Date: Wed, 13 Apr 2011 16:29:36 +0200 Subject: [PATCH] cpusets: randomize node rotor used in cpuset_mem_spread_node() [This patch has already been accepted as 0ac0c0d but later reverted (35926ff) because it itroduced arch specific __node_random which was defined only for x86 code so it broke other archs. This is a followup without any arch specific code. Other than that there are no functional changes.] Some workloads that create a large number of small files tend to assign too many pages to node 0 (multi-node systems). Part of the reason is that the rotor (in cpuset_mem_spread_node()) used to assign nodes starts at node 0 for newly created tasks. This patch changes the rotor to be initialized to a random node number of the cpuset. We are initializating it lazily in cpuset_mem_spread_node resp. cpuset_slab_spread_node. [akpm@linux-foundation.org: fix layout] [Lee.Schermerhorn@hp.com: Define stub numa_random() for !NUMA configuration] [mhocko@suse.cz: Make it arch independent + lazy initialization] Signed-off-by: Jack Steiner Signed-off-by: Lee Schermerhorn Signed-off-by: Michal Hocko Cc: Christoph Lameter Cc: Pekka Enberg Cc: Paul Menage Cc: Jack Steiner Cc: Robin Holt Cc: Andrew Morton Cc: Linus Torvalds --- include/linux/bitmap.h | 1 + include/linux/nodemask.h | 7 +++++++ kernel/cpuset.c | 8 ++++++++ kernel/fork.c | 5 +++++ lib/bitmap.c | 2 +- mm/mempolicy.c | 16 ++++++++++++++++ 6 files changed, 38 insertions(+), 1 deletions(-) diff --git a/include/linux/bitmap.h b/include/linux/bitmap.h index daf8c48..6fb2720 100644 --- a/include/linux/bitmap.h +++ b/include/linux/bitmap.h @@ -141,6 +141,7 @@ extern int bitmap_find_free_region(unsigned long *bitmap, int bits, int order); extern void bitmap_release_region(unsigned long *bitmap, int pos, int order); extern int bitmap_allocate_region(unsigned long *bitmap, int pos, int order); extern void bitmap_copy_le(void *dst, const unsigned long *src, int nbits); +extern int bitmap_ord_to_pos(const unsigned long *bitmap, int n, int bits); #define BITMAP_LAST_WORD_MASK(nbits) \ ( \ diff --git a/include/linux/nodemask.h b/include/linux/nodemask.h index dba35e4..66ccda9 100644 --- a/include/linux/nodemask.h +++ b/include/linux/nodemask.h @@ -66,6 +66,8 @@ * int num_online_nodes() Number of online Nodes * int num_possible_nodes() Number of all possible Nodes * + * int node_random(mask) Random node with set bit in mask + * * int node_online(node) Is some node online? * int node_possible(node) Is some node possible? * @@ -430,6 +432,9 @@ static inline void node_set_offline(int nid) node_clear_state(nid, N_ONLINE); nr_online_nodes = num_node_state(N_ONLINE); } + +extern int node_random(const nodemask_t *maskp); + #else static inline int node_state(int node, enum node_states state) @@ -460,6 +465,8 @@ static inline int num_node_state(enum node_states state) #define node_set_online(node) node_set_state((node), N_ONLINE) #define node_set_offline(node) node_clear_state((node), N_ONLINE) + +static inline int node_random(const nodemask_t *mask) { return 0; } #endif #define node_online_map node_states[N_ONLINE] diff --git a/kernel/cpuset.c b/kernel/cpuset.c index 33eee16..5376478 100644 --- a/kernel/cpuset.c +++ b/kernel/cpuset.c @@ -2465,11 +2465,19 @@ static int cpuset_spread_node(int *rotor) int cpuset_mem_spread_node(void) { + if (current->cpuset_mem_spread_rotor == -1) + current->cpuset_mem_spread_rotor = + node_random(¤t->mems_allowed); + return cpuset_spread_node(¤t->cpuset_mem_spread_rotor); } int cpuset_slab_spread_node(void) { + if (current->cpuset_slab_spread_rotor == -1) + current->cpuset_slab_spread_rotor + = node_random(¤t->mems_allowed); + return cpuset_spread_node(¤t->cpuset_slab_spread_rotor); } diff --git a/kernel/fork.c b/kernel/fork.c index e7548de..0883cff 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -1119,6 +1119,11 @@ static struct task_struct *copy_process(unsigned long clone_flags, } mpol_fix_fork_child_flag(p); #endif +#ifdef CONFIG_CPUSETS + /* Will be initialized lazily when first used */ + p->cpuset_mem_spread_rotor = -1; + p->cpuset_slab_spread_rotor = -1; +#endif #ifdef CONFIG_TRACE_IRQFLAGS p->irq_events = 0; #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW diff --git a/lib/bitmap.c b/lib/bitmap.c index 91e0ccf..a4b5217 100644 --- a/lib/bitmap.c +++ b/lib/bitmap.c @@ -671,7 +671,7 @@ static int bitmap_pos_to_ord(const unsigned long *buf, int pos, int bits) * * The bit positions 0 through @bits are valid positions in @buf. */ -static int bitmap_ord_to_pos(const unsigned long *buf, int ord, int bits) +int bitmap_ord_to_pos(const unsigned long *buf, int ord, int bits) { int pos = 0; diff --git a/mm/mempolicy.c b/mm/mempolicy.c index 959a8b8..8e57a72 100644 --- a/mm/mempolicy.c +++ b/mm/mempolicy.c @@ -93,6 +93,7 @@ #include #include +#include #include "internal.h" @@ -1649,6 +1650,21 @@ static inline unsigned interleave_nid(struct mempolicy *pol, return interleave_nodes(pol); } +/* + * Return the bit number of a random bit set in the nodemask. + * (returns -1 if nodemask is empty) + */ +int node_random(const nodemask_t *maskp) +{ + int w, bit = -1; + + w = nodes_weight(*maskp); + if (w) + bit = bitmap_ord_to_pos(maskp->bits, + get_random_int() % w, MAX_NUMNODES); + return bit; +} + #ifdef CONFIG_HUGETLBFS /* * huge_zonelist(@vma, @addr, @gfp_flags, @mpol) -- 1.7.4.1 -- Michal Hocko SUSE Labs SUSE LINUX s.r.o. Lihovarska 1060/12 190 00 Praha 9 Czech Republic -- 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/