Re: [PATCH bpf-next v2] bpf: Do not allocate percpu memory at init stage

[Alexei Starovoitov <alexei.starovoitov@xxxxxxxxx>]

On Fri, Nov 10, 2023 at 9:23 AM Yonghong Song <yonghong.song@xxxxxxxxx> wrote:
>
> +                               if (meta.func_id == special_kfunc_list[KF_bpf_percpu_obj_new_impl]) {
> +                                       if (!bpf_global_percpu_ma_set) {
> +                                               mutex_lock(&bpf_verifier_lock);
> +                                               if (!bpf_global_percpu_ma_set) {
> +                                                       err = bpf_mem_alloc_init(&bpf_global_percpu_ma, 0, true);
> +                                                       if (!err)
> +                                                               bpf_global_percpu_ma_set = true;
> +                                               }
> +                                               mutex_unlock(&bpf_verifier_lock);

I feel we're taking unnecessary risk here by reusing the mutex.
bpf_obj_new kfunc is a privileged operation and the verifier lock
is not held in such scenario, so it won't deadlock,
but let's just add another mutex to protect percpu_ma init.
Much easier to reason about.