Re: [PATCH] lockdep: slightly reorder 'struct lock_class' to save some memory

[Waiman Long]

On 11/26/23 03:56, Christophe JAILLET wrote:
> Based on pahole, 2 holes can be combined in the 'struct lock_class'. This
> saves 8 bytes in the structure on my x86_64.
>
> On a x86_64 configured with allmodconfig, this saves 64kb of memory in
> 'kernel/locking/lockdep.o', certainly because of:
>    struct lock_class lock_classes[MAX_LOCKDEP_KEYS];
>
> Before:
>     text	   data	    bss	    dec	    hex	filename
>   102501	1912490	11531636	13546627	 ceb483	kernel/locking/lockdep.o
>
> After:
>   102181	1912490	11466100	13480771	 cdb343	kernel/locking/lockdep.o
>
> Signed-off-by: Christophe JAILLET <christophe.jaillet@xxxxxxxxxx>
> ---
> After the reorder, pahole gives:
>
> struct lock_class {
>          struct hlist_node          hash_entry;           /*     0    16 */
>          struct list_head           lock_entry;           /*    16    16 */
>          struct list_head           locks_after;          /*    32    16 */
>          struct list_head           locks_before;         /*    48    16 */
>          /* --- cacheline 1 boundary (64 bytes) --- */
>          const struct lockdep_subclass_key  * key;        /*    64     8 */
>          lock_cmp_fn                cmp_fn;               /*    72     8 */
>          lock_print_fn              print_fn;             /*    80     8 */
>          unsigned int               subclass;             /*    88     4 */
>          unsigned int               dep_gen_id;           /*    92     4 */
>          long unsigned int          usage_mask;           /*    96     8 */
>          const struct lock_trace  * usage_traces[10];     /*   104    80 */
>          /* --- cacheline 2 boundary (128 bytes) was 56 bytes ago --- */
>          const char  *              name;                 /*   184     8 */
>          /* --- cacheline 3 boundary (192 bytes) --- */
>          int                        name_version;         /*   192     4 */
>          u8                         wait_type_inner;      /*   196     1 */
>          u8                         wait_type_outer;      /*   197     1 */
>          u8                         lock_type;            /*   198     1 */
>
>          /* XXX 1 byte hole, try to pack */
>
>          long unsigned int          contention_point[4];  /*   200    32 */
>          long unsigned int          contending_point[4];  /*   232    32 */
>
>          /* size: 264, cachelines: 5, members: 18 */
>          /* sum members: 263, holes: 1, sum holes: 1 */
>          /* last cacheline: 8 bytes */
> };
> ---
>   include/linux/lockdep_types.h | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/include/linux/lockdep_types.h b/include/linux/lockdep_types.h
> index 2ebc323d345a..857d785e89e6 100644
> --- a/include/linux/lockdep_types.h
> +++ b/include/linux/lockdep_types.h
> @@ -127,12 +127,12 @@ struct lock_class {
>   	unsigned long			usage_mask;
>   	const struct lock_trace		*usage_traces[LOCK_TRACE_STATES];
>   
> +	const char			*name;
>   	/*
>   	 * Generation counter, when doing certain classes of graph walking,
>   	 * to ensure that we check one node only once:
>   	 */
>   	int				name_version;
> -	const char			*name;
>   
>   	u8				wait_type_inner;
>   	u8				wait_type_outer;

LGTM

Acked-by: Waiman Long <longman@xxxxxxxxxx>