Re: [PATCH RESEND] fs: Move @f_count to different cacheline with @f_mode

[Jan Kara <jack@suse.cz>]

On Wed 24-06-20 16:32:28, Shaokun Zhang wrote:
> get_file_rcu_many, which is called by __fget_files, has used
> atomic_try_cmpxchg now and it can reduce the access number of the global
> variable to improve the performance of atomic instruction compared with
> atomic_cmpxchg. 
> 
> __fget_files does check the @f_mode with mask variable and will do some
> atomic operations on @f_count, but both are on the same cacheline.
> Many CPU cores do file access and it will cause much conflicts on @f_count. 
> If we could make the two members into different cachelines, it shall relax
> the siutations.

<snip nice unixbench results>

Thanks for the patch! The wins for your microbenchmark heavily sharing
struct file are nice but I'm not sure your change is a universal win. When
struct file is not shared (which is far more common), hot code paths like
__fget() or __fget_light() will now need to fetch two cache lines from
struct file instead of one. So I don't think that for most users the
tradeoff is really worth it...

								Honza

> diff --git a/include/linux/fs.h b/include/linux/fs.h
> index 3f881a892ea7..0faeab5622fb 100644
> --- a/include/linux/fs.h
> +++ b/include/linux/fs.h
> @@ -955,7 +955,6 @@ struct file {
>  	 */
>  	spinlock_t		f_lock;
>  	enum rw_hint		f_write_hint;
> -	atomic_long_t		f_count;
>  	unsigned int 		f_flags;
>  	fmode_t			f_mode;
>  	struct mutex		f_pos_lock;
> @@ -979,6 +978,7 @@ struct file {
>  	struct address_space	*f_mapping;
>  	errseq_t		f_wb_err;
>  	errseq_t		f_sb_err; /* for syncfs */
> +	atomic_long_t		f_count;
>  } __randomize_layout
>    __attribute__((aligned(4)));	/* lest something weird decides that 2 is OK */
>  
> -- 
> 2.7.4
> 
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR