[PATCH 4.9 155/206] f2fs: remove percpu_count due to performance regression
From: Greg Kroah-Hartman
Date: Tue Jan 10 2017 - 09:05:30 EST
4.9-stable review patch. If anyone has any objections, please let me know.
------------------
From: Jaegeuk Kim <jaegeuk@xxxxxxxxxx>
commit 35782b233f37e48ecc469d9c7232f3f6a7fad41a upstream.
This patch removes percpu_count usage due to performance regression in iozone.
Fixes: 523be8a6b3 ("f2fs: use percpu_counter for page counters")
Signed-off-by: Jaegeuk Kim <jaegeuk@xxxxxxxxxx>
Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
---
fs/f2fs/debug.c | 12 ++++++------
fs/f2fs/f2fs.h | 12 ++++++------
fs/f2fs/super.c | 16 +++++-----------
3 files changed, 17 insertions(+), 23 deletions(-)
--- a/fs/f2fs/debug.c
+++ b/fs/f2fs/debug.c
@@ -310,17 +310,17 @@ static int stat_show(struct seq_file *s,
seq_printf(s, " - Inner Struct Count: tree: %d(%d), node: %d\n",
si->ext_tree, si->zombie_tree, si->ext_node);
seq_puts(s, "\nBalancing F2FS Async:\n");
- seq_printf(s, " - inmem: %4lld, wb_bios: %4d\n",
+ seq_printf(s, " - inmem: %4d, wb_bios: %4d\n",
si->inmem_pages, si->wb_bios);
- seq_printf(s, " - nodes: %4lld in %4d\n",
+ seq_printf(s, " - nodes: %4d in %4d\n",
si->ndirty_node, si->node_pages);
- seq_printf(s, " - dents: %4lld in dirs:%4d (%4d)\n",
+ seq_printf(s, " - dents: %4d in dirs:%4d (%4d)\n",
si->ndirty_dent, si->ndirty_dirs, si->ndirty_all);
- seq_printf(s, " - datas: %4lld in files:%4d\n",
+ seq_printf(s, " - datas: %4d in files:%4d\n",
si->ndirty_data, si->ndirty_files);
- seq_printf(s, " - meta: %4lld in %4d\n",
+ seq_printf(s, " - meta: %4d in %4d\n",
si->ndirty_meta, si->meta_pages);
- seq_printf(s, " - imeta: %4lld\n",
+ seq_printf(s, " - imeta: %4d\n",
si->ndirty_imeta);
seq_printf(s, " - NATs: %9d/%9d\n - SITs: %9d/%9d\n",
si->dirty_nats, si->nats, si->dirty_sits, si->sits);
--- a/fs/f2fs/f2fs.h
+++ b/fs/f2fs/f2fs.h
@@ -819,7 +819,7 @@ struct f2fs_sb_info {
atomic_t nr_wb_bios; /* # of writeback bios */
/* # of pages, see count_type */
- struct percpu_counter nr_pages[NR_COUNT_TYPE];
+ atomic_t nr_pages[NR_COUNT_TYPE];
/* # of allocated blocks */
struct percpu_counter alloc_valid_block_count;
@@ -1233,7 +1233,7 @@ static inline void dec_valid_block_count
static inline void inc_page_count(struct f2fs_sb_info *sbi, int count_type)
{
- percpu_counter_inc(&sbi->nr_pages[count_type]);
+ atomic_inc(&sbi->nr_pages[count_type]);
if (count_type == F2FS_DIRTY_DATA || count_type == F2FS_INMEM_PAGES)
return;
@@ -1250,7 +1250,7 @@ static inline void inode_inc_dirty_pages
static inline void dec_page_count(struct f2fs_sb_info *sbi, int count_type)
{
- percpu_counter_dec(&sbi->nr_pages[count_type]);
+ atomic_dec(&sbi->nr_pages[count_type]);
}
static inline void inode_dec_dirty_pages(struct inode *inode)
@@ -1266,7 +1266,7 @@ static inline void inode_dec_dirty_pages
static inline s64 get_pages(struct f2fs_sb_info *sbi, int count_type)
{
- return percpu_counter_sum_positive(&sbi->nr_pages[count_type]);
+ return atomic_read(&sbi->nr_pages[count_type]);
}
static inline int get_dirty_pages(struct inode *inode)
@@ -2187,8 +2187,8 @@ struct f2fs_stat_info {
unsigned long long hit_largest, hit_cached, hit_rbtree;
unsigned long long hit_total, total_ext;
int ext_tree, zombie_tree, ext_node;
- s64 ndirty_node, ndirty_dent, ndirty_meta, ndirty_data, ndirty_imeta;
- s64 inmem_pages;
+ int ndirty_node, ndirty_dent, ndirty_meta, ndirty_data, ndirty_imeta;
+ int inmem_pages;
unsigned int ndirty_dirs, ndirty_files, ndirty_all;
int nats, dirty_nats, sits, dirty_sits, fnids;
int total_count, utilization;
--- a/fs/f2fs/super.c
+++ b/fs/f2fs/super.c
@@ -688,10 +688,6 @@ static void f2fs_destroy_inode(struct in
static void destroy_percpu_info(struct f2fs_sb_info *sbi)
{
- int i;
-
- for (i = 0; i < NR_COUNT_TYPE; i++)
- percpu_counter_destroy(&sbi->nr_pages[i]);
percpu_counter_destroy(&sbi->alloc_valid_block_count);
percpu_counter_destroy(&sbi->total_valid_inode_count);
}
@@ -1442,6 +1438,7 @@ int sanity_check_ckpt(struct f2fs_sb_inf
static void init_sb_info(struct f2fs_sb_info *sbi)
{
struct f2fs_super_block *raw_super = sbi->raw_super;
+ int i;
sbi->log_sectors_per_block =
le32_to_cpu(raw_super->log_sectors_per_block);
@@ -1466,6 +1463,9 @@ static void init_sb_info(struct f2fs_sb_
sbi->interval_time[REQ_TIME] = DEF_IDLE_INTERVAL;
clear_sbi_flag(sbi, SBI_NEED_FSCK);
+ for (i = 0; i < NR_COUNT_TYPE; i++)
+ atomic_set(&sbi->nr_pages[i], 0);
+
INIT_LIST_HEAD(&sbi->s_list);
mutex_init(&sbi->umount_mutex);
mutex_init(&sbi->wio_mutex[NODE]);
@@ -1481,13 +1481,7 @@ static void init_sb_info(struct f2fs_sb_
static int init_percpu_info(struct f2fs_sb_info *sbi)
{
- int i, err;
-
- for (i = 0; i < NR_COUNT_TYPE; i++) {
- err = percpu_counter_init(&sbi->nr_pages[i], 0, GFP_KERNEL);
- if (err)
- return err;
- }
+ int err;
err = percpu_counter_init(&sbi->alloc_valid_block_count, 0, GFP_KERNEL);
if (err)