On Fri, 2022-05-13 at 18:00 +0200, Daniel Borkmann wrote: > On 5/13/22 3:22 AM, Delyan Kratunov wrote: > [...] > > struct bpf_prog_array *bpf_prog_array_alloc(u32 prog_cnt, gfp_t flags); > > void bpf_prog_array_free(struct bpf_prog_array *progs); > > +/* Use when traversal over the bpf_prog_array uses tasks_trace rcu */ > > +void bpf_prog_array_free_sleepable(struct bpf_prog_array *progs); > > int bpf_prog_array_length(struct bpf_prog_array *progs); > > bool bpf_prog_array_is_empty(struct bpf_prog_array *array); > > int bpf_prog_array_copy_to_user(struct bpf_prog_array *progs, > > @@ -1451,6 +1454,56 @@ bpf_prog_run_array(const struct bpf_prog_array *array, > > return ret; > > } > > > > +/** > > + * Notes on RCU design for bpf_prog_arrays containing sleepable programs: > > + * > > + * We use the tasks_trace rcu flavor read section to protect the bpf_prog_array > > + * overall. As a result, we must use the bpf_prog_array_free_sleepable > > + * in order to use the tasks_trace rcu grace period. > > + * > > + * When a non-sleepable program is inside the array, we take the rcu read > > + * section and disable preemption for that program alone, so it can access > > + * rcu-protected dynamically sized maps. > > + */ > > Btw, there are a number of kdoc warnings around your series, pls make sure to > fix or use 'regular' comment: > > https://patchwork.hopto.org/static/nipa/641204/12848281/kdoc/stderr > https://patchwork.hopto.org/static/nipa/641204/12848282/kdoc/stderr Yeah, I just saw these too, I'll take care of them before the next reroll. Let's see if there's any other high level comments first. -- Delyan
