[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 <mhocko@xxxxxxx>
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@xxxxxxxxxxxxxxxxxxxx: fix layout]
[Lee.Schermerhorn@xxxxxx: Define stub numa_random() for !NUMA configuration]
[mhocko@xxxxxxx: Make it arch independent + lazy initialization]
Signed-off-by: Jack Steiner <steiner@xxxxxxx>
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@xxxxxx>
Signed-off-by: Michal Hocko <mhocko@xxxxxxx>
Cc: Christoph Lameter <cl@xxxxxxxxxxxxxxxxxxxx>
Cc: Pekka Enberg <penberg@xxxxxxxxxxxxxx>
Cc: Paul Menage <menage@xxxxxxxxxx>
Cc: Jack Steiner <steiner@xxxxxxx>
Cc: Robin Holt <holt@xxxxxxx>
Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
Cc: Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx>
---
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 <asm/tlbflush.h>
#include <asm/uaccess.h>
+#include <linux/random.h>
#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, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxxx For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>