Re: [PATCH bpf-next 1/2] bpf: Add bits iterator

[John Fastabend <john.fastabend@gmail.com>]

Alexei Starovoitov wrote:
> On Mon, Feb 26, 2024 at 3:21 PM John Fastabend <john.fastabend@gmail.com> wrote:
> > > +__bpf_kfunc int *bpf_iter_bits_next(struct bpf_iter_bits *it)
> > > +{
> > > +     struct bpf_iter_bits_kern *kit = (void *)it;
> > > +     const unsigned long *bits = kit->bits;
> > > +     int bit;
> > > +
> > > +     if (!bits)
> > > +             return NULL;
> > > +
> > > +     bit = find_next_bit(bits, kit->nr_bits, kit->bit + 1);
> >
> > Seems like this should be ok over unsafe memory as long as find_next_bit
> > is bounded?
> 
> Are you proposing to add find_next_bit() as a kfunc instead?

I was suggesting you could likely implement find_next_bit() in
BPF directly no need for a kfunc.

> 
> With the bpf_can_loop() proposal these two can be combined and
> it will probably achieve the same result.
> But imo this iterator is small enough to get in now and
> delete later when there is a better way.

Agree its fine to go in as is IMO. Mostly just curious if anyone
tried to implement it in BPF.

> Ideally we'd need to add new instructions to operate with bits efficiently.

+1.