On Mon, May 26, 2025 at 9:22 AM Leon Hwang <leon.hwang@xxxxxxxxx> wrote:
> +
> + data_sz = map->def.value_size;
> + if (is_percpu) {
> + num_cpus = libbpf_num_possible_cpus();
> + if (num_cpus < 0) {
> + err = num_cpus;
> + return err;
> + }
> +
> + data_sz = data_sz * num_cpus;
> + data = malloc(data_sz);
> + if (!data) {
> + err = -ENOMEM;
> + return err;
> + }
> +
> + elem_sz = map->def.value_size;
> + for (i = 0; i < num_cpus; i++)
> + memcpy(data + i * elem_sz, map->mmaped, elem_sz);
> + } else {
> + data = map->mmaped;
> + }
>
> if (obj->gen_loader) {
> bpf_gen__map_update_elem(obj->gen_loader, map - obj->maps,
> - map->mmaped, map->def.value_size);
> + data, data_sz);
I missed it earlier, but now I wonder how this is supposed to work ?
skel and lskel may be generated on a system with N cpus,
but loaded with M cpus.
Another concern is num_cpus multiplier can be huge.
lksel adds all that init data into a global array.
Pls avoid this multiplier.
