Re: [PATCH] bpf/verifier: optimize ID mapping reset in states_equal

[Eduard Zingerman <eddyz87@gmail.com>]

On Thu, 2026-01-15 at 22:49 +0800, wujing wrote:
> The verifier uses an ID mapping table (struct bpf_idmap) during state
> equivalence checks. Currently, reset_idmap_scratch performs a full memset
> on the entire map in every call.
> 
> The table size is exactly 4800 bytes (approx. 4.7KB), calculated as:
> - MAX_BPF_REG = 11
> - MAX_BPF_STACK = 512
> - BPF_REG_SIZE = 8
> - MAX_CALL_FRAMES = 8
> - BPF_ID_MAP_SIZE = (11 + 512 / 8) * 8 = 600 entries
> - Each entry (struct bpf_id_pair) is 8 bytes (two u32 fields)
> - Total size = 600 * 8 = 4800 bytes
> 
> For complex programs with many pruning points, this constant large memset
> introduces significant CPU overhead and cache pressure, especially when
> only a few IDs are actually used.
> 
> This patch optimizes the reset logic by:
> 1. Adding 'map_cnt' to bpf_idmap to track used slots.
> 2. Updating 'map_cnt' in check_ids to record the high-water mark.
> 3. Making reset_idmap_scratch perform a partial memset based on 'map_cnt'.
> 
> This improves pruning performance and reduces redundant memory writes.
> 
> Signed-off-by: wujing <realwujing@gmail.com>
                 ^^^^^^
		 Please use your full name.
> Signed-off-by: Qiliang Yuan <yuanql9@chinatelecom.cn>
> ---

I think this is an ok change.
Could you please collect some stats using 'perf stat' for some big selftest?

>  include/linux/bpf_verifier.h |  1 +
>  kernel/bpf/verifier.c        | 10 ++++++++--
>  2 files changed, 9 insertions(+), 2 deletions(-)
> 
> diff --git a/include/linux/bpf_verifier.h b/include/linux/bpf_verifier.h
> index 130bcbd66f60..562f7e63be29 100644
> --- a/include/linux/bpf_verifier.h
> +++ b/include/linux/bpf_verifier.h
> @@ -692,6 +692,7 @@ struct bpf_id_pair {
>  
>  struct bpf_idmap {
>  	u32 tmp_id_gen;
> +	u32 map_cnt;
>  	struct bpf_id_pair map[BPF_ID_MAP_SIZE];
>  };
>  
> diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
> index 37ce3990c9ad..6220dde41107 100644
> --- a/kernel/bpf/verifier.c
> +++ b/kernel/bpf/verifier.c
> @@ -18954,6 +18954,7 @@ static bool check_ids(u32 old_id, u32 cur_id, struct bpf_idmap *idmap)
>  			/* Reached an empty slot; haven't seen this id before */
>  			map[i].old = old_id;
>  			map[i].cur = cur_id;
> +			idmap->map_cnt = i + 1;
>  			return true;
>  		}
>  		if (map[i].old == old_id)
> @@ -19471,8 +19472,13 @@ static bool func_states_equal(struct bpf_verifier_env *env, struct bpf_func_stat
>  
>  static void reset_idmap_scratch(struct bpf_verifier_env *env)
>  {
> -	env->idmap_scratch.tmp_id_gen = env->id_gen;
> -	memset(&env->idmap_scratch.map, 0, sizeof(env->idmap_scratch.map));
> +	struct bpf_idmap *idmap = &env->idmap_scratch;
> +
> +	idmap->tmp_id_gen = env->id_gen;
> +	if (idmap->map_cnt) {

Nit: this condition is not really necessary.

> +		memset(idmap->map, 0, idmap->map_cnt * sizeof(struct bpf_id_pair));
> +		idmap->map_cnt = 0;
> +	}
>  }
>  
>  static bool states_equal(struct bpf_verifier_env *env,