Re: [linus:master] [file] 0ede61d858: will-it-scale.per_thread_ops -2.9% regression

[Mateusz Guzik]

On Mon, Nov 20, 2023 at 03:11:31PM +0800, kernel test robot wrote:
> 
> 
> Hello,
> 
> kernel test robot noticed a -2.9% regression of will-it-scale.per_thread_ops on:
> 
> 
> commit: 0ede61d8589cc2d93aa78230d74ac58b5b8d0244 ("file: convert to SLAB_TYPESAFE_BY_RCU")
> https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git master
> 
> 93faf426e3cc000c 0ede61d8589cc2d93aa78230d74 
> ---------------- --------------------------- 
>          %stddev     %change         %stddev
>              \          |                \  
[snip]
>      30.90 ±  4%     -20.6       10.35 ±  2%  perf-profile.self.cycles-pp.__fget_light
>       0.00           +26.5       26.48        perf-profile.self.cycles-pp.__get_file_rcu
[snip]

So __fget_light now got a func call.

I don't know if this is worth patching (and benchmarking after), but I
if sorting this out is of interest, triviality below is probably the
easiest way out:

diff --git a/fs/file.c b/fs/file.c
index 5fb0b146e79e..d8d3e18800c4 100644
--- a/fs/file.c
+++ b/fs/file.c
@@ -856,14 +856,14 @@ void do_close_on_exec(struct files_struct *files)
 	spin_unlock(&files->file_lock);
 }
 
-static struct file *__get_file_rcu(struct file __rcu **f)
+static __always_inline struct file *__get_file_rcu(struct file __rcu **f)
 {
 	struct file __rcu *file;
 	struct file __rcu *file_reloaded;
 	struct file __rcu *file_reloaded_cmp;
 
 	file = rcu_dereference_raw(*f);
-	if (!file)
+	if (unlikely(!file))
 		return NULL;
 
 	if (unlikely(!atomic_long_inc_not_zero(&file->f_count)))
@@ -891,7 +891,7 @@ static struct file *__get_file_rcu(struct file __rcu **f)
 	 * If the pointers don't match the file has been reallocated by
 	 * SLAB_TYPESAFE_BY_RCU.
 	 */
-	if (file == file_reloaded_cmp)
+	if (likely(file == file_reloaded_cmp))
 		return file_reloaded;
 
 	fput(file);