Re: [PATCH bpf-next 2/3] libbpf: split bpf object load into prepare/load

[Eduard Zingerman <eddyz87@gmail.com>]

On Mon, 2025-03-03 at 13:38 -0800, Andrii Nakryiko wrote:
> On Sat, Mar 1, 2025 at 1:45 PM Mykyta Yatsenko
> <mykyta.yatsenko5@gmail.com> wrote:
> > 
> > On 01/03/2025 08:12, Eduard Zingerman wrote:
> > > On Fri, 2025-02-28 at 17:52 +0000, Mykyta Yatsenko wrote:
> > > 
> > > [...]
> > > 
> > > > diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c
> > > > index 9ced1ce2334c..dd2f64903c3b 100644
> > > > --- a/tools/lib/bpf/libbpf.c
> > > > +++ b/tools/lib/bpf/libbpf.c
> > > > @@ -4858,7 +4858,7 @@ bool bpf_map__autocreate(const struct bpf_map *map)
> > > > 
> > > >   int bpf_map__set_autocreate(struct bpf_map *map, bool autocreate)
> > > >   {
> > > > -    if (map->obj->state >= OBJ_LOADED)
> > > > +    if (map->obj->state >= OBJ_PREPARED)
> > > >              return libbpf_err(-EBUSY);
> > > I looked through logic in patches #1 and #2 and changes look correct.
> > > Running tests under valgrind does not show issues with this feature.
> > > The only ask from my side is to consider doing ==/!= comparisons in
> > > cases like above. E.g. it seems that `map->obj->state != OBJ_OPENED`
> > > is a bit simpler to understand when reading condition above.
> > > Or maybe that's just me.
> > I'm not sure about this one.  >= or < checks for state relative to
> > operand more
> > flexibly,for example `map->obj->state >= OBJ_PREPARED` is read as
> > "is the object in at least PREPARED state". Perhaps, if we add more states,
> > these >,< checks will not require any changes, while ==, != may.
> > I guess this also depends on what we actually want to check here, is it that
> > state at least PREPARED or the state is not initial OPENED.
> > Not a strong opinion, though, happy to flip code to ==, !=.
> 
> Those steps are logically ordered, so >= and <= makes more sense. If
> we ever add one extra step somewhere in between existing steps, most
> checks will stay correct, while with equality a lot of them might need
> to be adjusted to multiple equalities.

As I said, for me personally it is easier to read "can set autocreate
only in OPENED state", compared to "can't set autocreate if state is
PREPARED of higher".
But whatever, I'm not a true C programmer anyway :)

[...]