Please ignore this patch. Send out the old version that has not been
proofread yet. Will resend. Sorry for the inconvenience.
On 3/15/2025 11:02 PM, Hou Tao wrote:
> From: Hou Tao <houtao1@xxxxxxxxxx>
>
> When there is no special field in the map value, there is no need to
> invoke bpf_obj_free_fields(). Therefore, checking the validity of
> map->record in advance.
>
> After the change, the benchmark result of per-cpu update case in
> map_perf_test increase 30%+.
>
> Signed-off-by: Hou Tao <houtao1@xxxxxxxxxx>
> ---
> kernel/bpf/hashtab.c | 3 +++
> 1 file changed, 3 insertions(+)
>
> diff --git a/kernel/bpf/hashtab.c b/kernel/bpf/hashtab.c
> index 6527e9ce83cd9..2623970175dcf 100644
> --- a/kernel/bpf/hashtab.c
> +++ b/kernel/bpf/hashtab.c
> @@ -906,6 +906,9 @@ static void htab_free_dynptr_key(struct bpf_htab *htab, void *key)
> static void check_and_free_fields(struct bpf_htab *htab,
> struct htab_elem *elem)
> {
> + if (IS_ERR_OR_NULL(htab->map.record))
> + return;
> +
> if (htab_is_percpu(htab)) {
> void __percpu *pptr = htab_elem_get_ptr(elem, htab->map.key_size);
> int cpu;
