vfs: remove redundant smp_mb for thp handling in do_dentry_open

opening for write performs:

if (f->f_mode & FMODE_WRITE) {
[snip]
        smp_mb();
        if (filemap_nr_thps(inode->i_mapping)) {
[snip]
        }
}

filemap_nr_thps on kernels built without CONFIG_READ_ONLY_THP_FOR
expands to 0, allowing the compiler to eliminate the entire thing, with
exception of the fence (and the branch leading there).

So happens required synchronisation between i_writecount and nr_thps
changes is already provided by the full fence coming from
get_write_access -> atomic_inc_unless_negative, thus the smp_mb instance
above can be removed regardless of CONFIG_READ_ONLY_THP_FOR.

While I updated commentary in places claiming to match the now-removed
fence, I did not try to patch them to act on the compile option.

I did not bother benchmarking it, not issuing a spurious full fence in
the fast path does not warrant justification from perf standpoint.

Signed-off-by: Mateusz Guzik <mjguzik@gmail.com>
Link: https://lore.kernel.org/r/20240624085402.493630-1-mjguzik@gmail.com
Signed-off-by: Christian Brauner <brauner@kernel.org>

1 files changed, 4 insertions, 5 deletions

diff --git a/fs/open.c b/fs/open.c
index a5c4f8a0f143..c4e9b01aafd8 100644
--- a/fs/open.c
+++ b/fs/open.c
@@ -986,12 +986,11 @@ static int do_dentry_open(struct file *f,
 	 */
 	if (f->f_mode & FMODE_WRITE) {
 		/*
-		 * Paired with smp_mb() in collapse_file() to ensure nr_thps
-		 * is up to date and the update to i_writecount by
-		 * get_write_access() is visible. Ensures subsequent insertion
-		 * of THPs into the page cache will fail.
+		 * Depends on full fence from get_write_access() to synchronize
+		 * against collapse_file() regarding i_writecount and nr_thps
+		 * updates. Ensures subsequent insertion of THPs into the page
+		 * cache will fail.
 		 */
-		smp_mb();
 		if (filemap_nr_thps(inode->i_mapping)) {
 			struct address_space *mapping = inode->i_mapping;