I discovered on oprofile hunting on a SMP platform that dentry lookups were
slowed down because d_hash_mask, d_hash_shift and dentry_hashtable were in a
cache line that contained inodes_stat. So each time inodes_stats is changed by
a cpu, other cpus have to refill their cache line.
This patch moves some variables to the __read_mostly section, in order to
avoid false sharing. RCU dentry lookups can go full speed.
Before someone asks, it is valid to declare a pointer as 'read mostly', even
if the data pointed by the pointer is heavily modified. hash table pointers
and kmem_cache pointers are setup at boot time, so they are perfect candidates
to 'read_mostly' section. Same apply for 'struct vfsmount *'
Signed-off-by: Eric Dumazet <[email protected]>
--- a/fs/dcache.c 2006-03-21 13:48:13.000000000 +0100
+++ b/fs/dcache.c 2006-03-21 13:55:00.000000000 +0100
@@ -36,7 +36,7 @@
/* #define DCACHE_DEBUG 1 */
-int sysctl_vfs_cache_pressure = 100;
+int sysctl_vfs_cache_pressure __read_mostly = 100;
EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
__cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lock);
@@ -44,7 +44,7 @@
EXPORT_SYMBOL(dcache_lock);
-static kmem_cache_t *dentry_cache;
+static kmem_cache_t *dentry_cache __read_mostly;
#define DNAME_INLINE_LEN (sizeof(struct dentry)-offsetof(struct dentry,d_iname))
@@ -59,9 +59,9 @@
#define D_HASHBITS d_hash_shift
#define D_HASHMASK d_hash_mask
-static unsigned int d_hash_mask;
-static unsigned int d_hash_shift;
-static struct hlist_head *dentry_hashtable;
+static unsigned int d_hash_mask __read_mostly;
+static unsigned int d_hash_shift __read_mostly;
+static struct hlist_head *dentry_hashtable __read_mostly;
static LIST_HEAD(dentry_unused);
/* Statistics gathering. */
@@ -1706,10 +1706,10 @@
}
/* SLAB cache for __getname() consumers */
-kmem_cache_t *names_cachep;
+kmem_cache_t *names_cachep __read_mostly;
/* SLAB cache for file structures */
-kmem_cache_t *filp_cachep;
+kmem_cache_t *filp_cachep __read_mostly;
EXPORT_SYMBOL(d_genocide);
--- a/fs/inode.c 2006-03-21 13:50:19.000000000 +0100
+++ b/fs/inode.c 2006-03-21 13:54:39.000000000 +0100
@@ -56,8 +56,8 @@
#define I_HASHBITS i_hash_shift
#define I_HASHMASK i_hash_mask
-static unsigned int i_hash_mask;
-static unsigned int i_hash_shift;
+static unsigned int i_hash_mask __read_mostly;
+static unsigned int i_hash_shift __read_mostly;
/*
* Each inode can be on two separate lists. One is
@@ -73,7 +73,7 @@
LIST_HEAD(inode_in_use);
LIST_HEAD(inode_unused);
-static struct hlist_head *inode_hashtable;
+static struct hlist_head *inode_hashtable __read_mostly;
/*
* A simple spinlock to protect the list manipulations.
@@ -98,7 +98,7 @@
*/
struct inodes_stat_t inodes_stat;
-static kmem_cache_t * inode_cachep;
+static kmem_cache_t * inode_cachep __read_mostly;
static struct inode *alloc_inode(struct super_block *sb)
{
--- a/fs/namespace.c 2006-03-22 05:20:33.000000000 +0100
+++ b/fs/namespace.c 2006-03-22 05:23:40.000000000 +0100
@@ -43,9 +43,9 @@
static int event;
-static struct list_head *mount_hashtable;
+static struct list_head *mount_hashtable __read_mostly;
static int hash_mask __read_mostly, hash_bits __read_mostly;
-static kmem_cache_t *mnt_cache;
+static kmem_cache_t *mnt_cache __read_mostly;
static struct rw_semaphore namespace_sem;
/* /sys/fs */
--- a/fs/dcookies.c 2006-03-22 05:35:46.000000000 +0100
+++ b/fs/dcookies.c 2006-03-22 05:36:55.000000000 +0100
@@ -37,9 +37,9 @@
static LIST_HEAD(dcookie_users);
static DECLARE_MUTEX(dcookie_sem);
-static kmem_cache_t * dcookie_cache;
-static struct list_head * dcookie_hashtable;
-static size_t hash_size;
+static kmem_cache_t * dcookie_cache __read_mostly;
+static struct list_head * dcookie_hashtable __read_mostly;
+static size_t hash_size __read_mostly;
static inline int is_live(void)
{
--- a/fs/fcntl.c 2006-03-22 05:37:40.000000000 +0100
+++ b/fs/fcntl.c 2006-03-22 05:43:49.000000000 +0100
@@ -413,7 +413,7 @@
/* Table to convert sigio signal codes into poll band bitmaps */
-static long band_table[NSIGPOLL] = {
+static const long band_table[NSIGPOLL] = {
POLLIN | POLLRDNORM, /* POLL_IN */
POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
@@ -532,7 +532,7 @@
}
static DEFINE_RWLOCK(fasync_lock);
-static kmem_cache_t *fasync_cache;
+static kmem_cache_t *fasync_cache __read_mostly;
/*
* fasync_helper() is used by some character device drivers (mainly mice)
--- a/fs/eventpoll.c 2006-03-22 05:39:06.000000000 +0100
+++ b/fs/eventpoll.c 2006-03-22 05:43:49.000000000 +0100
@@ -280,13 +280,13 @@
static struct poll_safewake psw;
/* Slab cache used to allocate "struct epitem" */
-static kmem_cache_t *epi_cache;
+static kmem_cache_t *epi_cache __read_mostly;
/* Slab cache used to allocate "struct eppoll_entry" */
-static kmem_cache_t *pwq_cache;
+static kmem_cache_t *pwq_cache __read_mostly;
/* Virtual fs used to allocate inodes for eventpoll files */
-static struct vfsmount *eventpoll_mnt;
+static struct vfsmount *eventpoll_mnt __read_mostly;
/* File callbacks that implement the eventpoll file behaviour */
static struct file_operations eventpoll_fops = {
--- a/fs/inotify.c 2006-03-22 05:40:44.000000000 +0100
+++ b/fs/inotify.c 2006-03-22 05:43:49.000000000 +0100
@@ -40,15 +40,15 @@
static atomic_t inotify_cookie;
static atomic_t inotify_watches;
-static kmem_cache_t *watch_cachep;
-static kmem_cache_t *event_cachep;
+static kmem_cache_t *watch_cachep __read_mostly;
+static kmem_cache_t *event_cachep __read_mostly;
-static struct vfsmount *inotify_mnt;
+static struct vfsmount *inotify_mnt __read_mostly;
/* these are configurable via /proc/sys/fs/inotify/ */
-int inotify_max_user_instances;
-int inotify_max_user_watches;
-int inotify_max_queued_events;
+int inotify_max_user_instances __read_mostly;
+int inotify_max_user_watches __read_mostly;
+int inotify_max_queued_events __read_mostly;
/*
* Lock ordering:
--- a/fs/locks.c 2006-03-22 05:43:34.000000000 +0100
+++ b/fs/locks.c 2006-03-22 05:43:49.000000000 +0100
@@ -145,7 +145,7 @@
static LIST_HEAD(blocked_list);
-static kmem_cache_t *filelock_cache;
+static kmem_cache_t *filelock_cache __read_mostly;
/* Allocate an empty lock structure. */
static struct file_lock *locks_alloc_lock(void)
--- a/fs/bio.c 2006-03-22 05:44:20.000000000 +0100
+++ b/fs/bio.c 2006-03-22 05:50:37.000000000 +0100
@@ -29,7 +29,7 @@
#define BIO_POOL_SIZE 256
-static kmem_cache_t *bio_slab;
+static kmem_cache_t *bio_slab __read_mostly;
#define BIOVEC_NR_POOLS 6
@@ -38,7 +38,7 @@
* basically we just need to survive
*/
#define BIO_SPLIT_ENTRIES 8
-mempool_t *bio_split_pool;
+mempool_t *bio_split_pool __read_mostly;
struct biovec_slab {
int nr_vecs;
--- a/fs/dnotify.c 2006-03-22 05:46:33.000000000 +0100
+++ b/fs/dnotify.c 2006-03-22 05:50:37.000000000 +0100
@@ -21,9 +21,9 @@
#include <linux/spinlock.h>
#include <linux/slab.h>
-int dir_notify_enable = 1;
+int dir_notify_enable __read_mostly = 1;
-static kmem_cache_t *dn_cache;
+static kmem_cache_t *dn_cache __read_mostly;
static void redo_inode_mask(struct inode *inode)
{
--- a/fs/block_dev.c 2006-03-22 05:48:29.000000000 +0100
+++ b/fs/block_dev.c 2006-03-22 05:50:37.000000000 +0100
@@ -238,7 +238,7 @@
*/
static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
-static kmem_cache_t * bdev_cachep;
+static kmem_cache_t * bdev_cachep __read_mostly;
static struct inode *bdev_alloc_inode(struct super_block *sb)
{
@@ -312,7 +312,7 @@
.kill_sb = kill_anon_super,
};
-static struct vfsmount *bd_mnt;
+static struct vfsmount *bd_mnt __read_mostly;
struct super_block *blockdev_superblock;
void __init bdev_cache_init(void)
--- a/fs/pipe.c 2006-03-22 05:51:16.000000000 +0100
+++ b/fs/pipe.c 2006-03-22 05:54:11.000000000 +0100
@@ -676,7 +676,7 @@
return NULL;
}
-static struct vfsmount *pipe_mnt;
+static struct vfsmount *pipe_mnt __read_mostly;
static int pipefs_delete_dentry(struct dentry *dentry)
{
return 1;
[Index of Archives]
[Kernel Newbies]
[Netfilter]
[Bugtraq]
[Photo]
[Stuff]
[Gimp]
[Yosemite News]
[MIPS Linux]
[ARM Linux]
[Linux Security]
[Linux RAID]
[Video 4 Linux]
[Linux for the blind]
[Linux Resources]
