The patch titled tmpfs: make tmpfs scalable with caches for free blocks has been added to the -mm tree. Its filename is tmpfs-make-tmpfs-scalable-with-caches-for-free-blocks.patch Before you just go and hit "reply", please: a) Consider who else should be cc'ed b) Prefer to cc a suitable mailing list as well c) Ideally: find the original patch on the mailing list and do a reply-to-all to that, adding suitable additional cc's *** Remember to use Documentation/SubmitChecklist when testing your code *** See http://userweb.kernel.org/~akpm/stuff/added-to-mm.txt to find out what to do about this The current -mm tree may be found at http://userweb.kernel.org/~akpm/mmotm/ ------------------------------------------------------ Subject: tmpfs: make tmpfs scalable with caches for free blocks From: Tim Chen <tim.c.chen@xxxxxxxxxxxxxxx> The current implementation of tmpfs is not scalable. The stat_lock is contended whenever we need to get a new page, leading to lots lock contentions. This patch makes use of the qtoken library to maintain local caches of free pages to speed up getting and returning of pages without acquisition of stat_lock improve the performance of tmpfs. Signed-off-by: Tim Chen <tim.c.chen@xxxxxxxxxxxxxxx> Cc: Andi Kleen <ak@xxxxxxxxxxxxxxx> Cc: Hugh Dickins <hughd@xxxxxxxxxx> Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> --- include/linux/shmem_fs.h | 4 +- mm/shmem.c | 58 ++++++++++++++++++++----------------- 2 files changed, 35 insertions(+), 27 deletions(-) diff -puN include/linux/shmem_fs.h~tmpfs-make-tmpfs-scalable-with-caches-for-free-blocks include/linux/shmem_fs.h --- a/include/linux/shmem_fs.h~tmpfs-make-tmpfs-scalable-with-caches-for-free-blocks +++ a/include/linux/shmem_fs.h @@ -3,10 +3,12 @@ #include <linux/swap.h> #include <linux/mempolicy.h> +#include <linux/qtoken.h> /* inode in-kernel data */ #define SHMEM_NR_DIRECT 16 +#define SHMEM_FREE_BLK_CACHE_SZ 512 struct shmem_inode_info { spinlock_t lock; @@ -23,7 +25,7 @@ struct shmem_inode_info { struct shmem_sb_info { unsigned long max_blocks; /* How many blocks are allowed */ - unsigned long free_blocks; /* How many are left for allocation */ + struct qtoken token_jar; /* Token jar of free blocks */ unsigned long max_inodes; /* How many inodes are allowed */ unsigned long free_inodes; /* How many are left for allocation */ spinlock_t stat_lock; /* Serialize shmem_sb_info changes */ diff -puN mm/shmem.c~tmpfs-make-tmpfs-scalable-with-caches-for-free-blocks mm/shmem.c --- a/mm/shmem.c~tmpfs-make-tmpfs-scalable-with-caches-for-free-blocks +++ a/mm/shmem.c @@ -29,6 +29,7 @@ #include <linux/mm.h> #include <linux/module.h> #include <linux/swap.h> +#include <linux/qtoken.h> static struct vfsmount *shm_mnt; @@ -233,10 +234,10 @@ static void shmem_free_blocks(struct ino { struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); if (sbinfo->max_blocks) { - spin_lock(&sbinfo->stat_lock); - sbinfo->free_blocks += pages; + spin_lock(&inode->i_lock); + qtoken_return(&sbinfo->token_jar, pages); inode->i_blocks -= pages*BLOCKS_PER_PAGE; - spin_unlock(&sbinfo->stat_lock); + spin_unlock(&inode->i_lock); } } @@ -416,19 +417,18 @@ static swp_entry_t *shmem_swp_alloc(stru if (sgp == SGP_READ) return shmem_swp_map(ZERO_PAGE(0)); /* - * Test free_blocks against 1 not 0, since we have 1 data - * page (and perhaps indirect index pages) yet to allocate: + * Leave 1 pg in reserve in token jar, since we have + * 1 data pg (and perhaps indirect index pages) yet to allocate: * a waste to allocate index if we cannot allocate data. */ if (sbinfo->max_blocks) { - spin_lock(&sbinfo->stat_lock); - if (sbinfo->free_blocks <= 1) { - spin_unlock(&sbinfo->stat_lock); + spin_lock(&inode->i_lock); + if (qtoken_get(&sbinfo->token_jar, 1, 1)) { + spin_unlock(&inode->i_lock); return ERR_PTR(-ENOSPC); } - sbinfo->free_blocks--; inode->i_blocks += BLOCKS_PER_PAGE; - spin_unlock(&sbinfo->stat_lock); + spin_unlock(&inode->i_lock); } spin_unlock(&info->lock); @@ -1386,17 +1386,20 @@ repeat: shmem_swp_unmap(entry); sbinfo = SHMEM_SB(inode->i_sb); if (sbinfo->max_blocks) { - spin_lock(&sbinfo->stat_lock); - if (sbinfo->free_blocks == 0 || - shmem_acct_block(info->flags)) { - spin_unlock(&sbinfo->stat_lock); + if (shmem_acct_block(info->flags)) { + spin_unlock(&info->lock); + error = -ENOSPC; + goto failed; + } + spin_lock(&inode->i_lock); + if (qtoken_get(&sbinfo->token_jar, 1, 0)) { + spin_unlock(&inode->i_lock); spin_unlock(&info->lock); error = -ENOSPC; goto failed; } - sbinfo->free_blocks--; inode->i_blocks += BLOCKS_PER_PAGE; - spin_unlock(&sbinfo->stat_lock); + spin_unlock(&inode->i_lock); } else if (shmem_acct_block(info->flags)) { spin_unlock(&info->lock); error = -ENOSPC; @@ -1793,7 +1796,7 @@ static int shmem_statfs(struct dentry *d spin_lock(&sbinfo->stat_lock); if (sbinfo->max_blocks) { buf->f_blocks = sbinfo->max_blocks; - buf->f_bavail = buf->f_bfree = sbinfo->free_blocks; + buf->f_bavail = buf->f_bfree = qtoken_avail(&sbinfo->token_jar); } if (sbinfo->max_inodes) { buf->f_files = sbinfo->max_inodes; @@ -2241,7 +2244,6 @@ static int shmem_remount_fs(struct super { struct shmem_sb_info *sbinfo = SHMEM_SB(sb); struct shmem_sb_info config = *sbinfo; - unsigned long blocks; unsigned long inodes; int error = -EINVAL; @@ -2249,12 +2251,6 @@ static int shmem_remount_fs(struct super return error; spin_lock(&sbinfo->stat_lock); - blocks = sbinfo->max_blocks - sbinfo->free_blocks; - inodes = sbinfo->max_inodes - sbinfo->free_inodes; - if (config.max_blocks < blocks) - goto out; - if (config.max_inodes < inodes) - goto out; /* * Those tests also disallow limited->unlimited while any are in * use, so i_blocks will always be zero when max_blocks is zero; @@ -2265,10 +2261,14 @@ static int shmem_remount_fs(struct super goto out; if (config.max_inodes && !sbinfo->max_inodes) goto out; + inodes = sbinfo->max_inodes - sbinfo->free_inodes; + if (config.max_inodes < inodes) + goto out; + if (qtoken_resize(&sbinfo->token_jar, config.max_blocks)) + goto out; error = 0; sbinfo->max_blocks = config.max_blocks; - sbinfo->free_blocks = config.max_blocks - blocks; sbinfo->max_inodes = config.max_inodes; sbinfo->free_inodes = config.max_inodes - inodes; @@ -2301,6 +2301,10 @@ static int shmem_show_options(struct seq static void shmem_put_super(struct super_block *sb) { + struct shmem_sb_info *sbinfo = SHMEM_SB(sb); + + if (sbinfo) + qtoken_free(&sbinfo->token_jar); kfree(sb->s_fs_info); sb->s_fs_info = NULL; } @@ -2343,8 +2347,10 @@ int shmem_fill_super(struct super_block #endif spin_lock_init(&sbinfo->stat_lock); - sbinfo->free_blocks = sbinfo->max_blocks; sbinfo->free_inodes = sbinfo->max_inodes; + if (qtoken_init(&sbinfo->token_jar, sbinfo->max_blocks, + SHMEM_FREE_BLK_CACHE_SZ)) + goto failed; sb->s_maxbytes = SHMEM_MAX_BYTES; sb->s_blocksize = PAGE_CACHE_SIZE; _ Patches currently in -mm which might be from tim.c.chen@xxxxxxxxxxxxxxx are tmpfs-quick-token-library-to-allow-scalable-retrieval-of-tokens-from-token-jar.patch tmpfs-quick-token-library-to-allow-scalable-retrieval-of-tokens-from-token-jar-fix.patch tmpfs-make-tmpfs-scalable-with-caches-for-free-blocks.patch -- To unsubscribe from this list: send the line "unsubscribe mm-commits" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html