Re: [PATCH bpf-next v1 1/8] bpf: Introduce pseudo_btf_id

[Alexei Starovoitov]

On Wed, Aug 19, 2020 at 03:40:23PM -0700, Hao Luo wrote:
> +
>  /* verify BPF_LD_IMM64 instruction */
>  static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn)
>  {
> @@ -7234,6 +7296,9 @@ static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn)
>  		return 0;
>  	}
>  
> +	if (insn->src_reg == BPF_PSEUDO_BTF_ID)
> +		return check_pseudo_btf_id(env, insn);
> +
>  	map = env->used_maps[aux->map_index];
>  	mark_reg_known_zero(env, regs, insn->dst_reg);
>  	regs[insn->dst_reg].map_ptr = map;
> @@ -9255,6 +9320,9 @@ static int replace_map_fd_with_map_ptr(struct bpf_verifier_env *env)
>  				/* valid generic load 64-bit imm */
>  				goto next_insn;
>  
> +			if (insn[0].src_reg == BPF_PSEUDO_BTF_ID)
> +				goto next_insn;
> +

Why did you choose to do it during main do_check() walk instead of this pre-pass ?
check_ld_imm() can be called multiple times for the same insn,
so it's faster and less surprising to do it during replace_map_fd_with_map_ptr().
BTF needs to be parsed first, of course.
You can either move check_btf_info() before replace_map_fd_with_map_ptr() or
move replace_map_fd_with_map_ptr() after check_btf_info().
The latter is probably cleaner.