fs: group frequently accessed fields of struct super_block together

Kernel test robot reported [1] a 6% performance regression in a
concurrent unlink(2) workload on commit 60f7ed8c7c4d ("fsnotify: send
path type events to group with super block marks").

The performance test was run with no fsnotify marks at all on the
data set, so the only extra instructions added by the offending
commit are tests of the super_block fields s_fsnotify_{marks,mask}
and these tests happen on almost every single inode access.

When adding those fields to the super_block struct, we did not give much
thought of placing them on a hot cache lines (we just placed them at the
end of the struct).

Re-organize struct super_block to try and keep some frequently accessed
fields on the same cache line.

Move the frequently accessed fields s_fsnotify_{marks,mask} near the
frequently accessed fields s_fs_info,s_time_gran, while filling a 64bit
alignment hole after s_time_gran.

Move the seldom accessed fields s_id,s_uuid,s_max_links,s_mode near the
seldom accessed fields s_vfs_rename_mutex,s_subtype.

Rong Chen confirmed that this patch solved the reported problem.

[1] https://lkml.org/lkml/2018/9/30/206

Reported-by: kernel test robot <rong.a.chen@intel.com>
Tested-by: kernel test robot <rong.a.chen@intel.com>
Fixes: 1e6cb72399 ("fsnotify: add super block object type")
Signed-off-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Jan Kara <jack@suse.cz>

diff --git a/include/linux/fs.h b/include/linux/fs.h
index 2c14801d0aa3e64730870dc002014ab83c77ab3e..6da94deb957ff7314f92285b4e1dad59deee45e2 100644 (file)
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -1393,17 +1393,26 @@ struct super_block {
 
        struct sb_writers       s_writers;
 
+       /*
+        * Keep s_fs_info, s_time_gran, s_fsnotify_mask, and
+        * s_fsnotify_marks together for cache efficiency. They are frequently
+        * accessed and rarely modified.
+        */
+       void                    *s_fs_info;     /* Filesystem private info */
+
+       /* Granularity of c/m/atime in ns (cannot be worse than a second) */
+       u32                     s_time_gran;
+#ifdef CONFIG_FSNOTIFY
+       __u32                   s_fsnotify_mask;
+       struct fsnotify_mark_connector __rcu    *s_fsnotify_marks;
+#endif
+
        char                    s_id[32];       /* Informational name */
        uuid_t                  s_uuid;         /* UUID */
 
-       void                    *s_fs_info;     /* Filesystem private info */
        unsigned int            s_max_links;
        fmode_t                 s_mode;
 
-       /* Granularity of c/m/atime in ns.
-          Cannot be worse than a second */
-       u32                s_time_gran;
-
        /*
         * The next field is for VFS *only*. No filesystems have any business
         * even looking at it. You had been warned.
@@ -1464,11 +1473,6 @@ struct super_block {
 
        spinlock_t              s_inode_wblist_lock;
        struct list_head        s_inodes_wb;    /* writeback inodes */
-
-#ifdef CONFIG_FSNOTIFY
-       __u32                   s_fsnotify_mask;
-       struct fsnotify_mark_connector __rcu    *s_fsnotify_marks;
-#endif
 } __randomize_layout;
 
 /* Helper functions so that in most cases filesystems will