On Tue, Nov 30, 2021 at 3:06 PM Alexei Starovoitov
<alexei.starovoitov@xxxxxxxxx> wrote:
>
> On Mon, Nov 29, 2021 at 8:10 PM Andrii Nakryiko
> <andrii.nakryiko@xxxxxxxxx> wrote:
> >
> > oh, I thought you added those fields initially and forgot to delete or
> > something, didn't notice that you are just "opting them out" for
> > __KERNEL__. I think libbpf code doesn't strictly need this, here's the
> > diff that completely removes their use, it's pretty straightforward
> > and minimal, so maybe instead of #ifdef'ing let's just do that?
> >
> > diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c
> > index b59fede08ba7..95fa57eea289 100644
> > --- a/tools/lib/bpf/libbpf.c
> > +++ b/tools/lib/bpf/libbpf.c
> > @@ -5179,15 +5179,18 @@ static int bpf_core_add_cands(struct
> > bpf_core_cand *local_cand,
> > struct bpf_core_cand_list *cands)
> > {
> > struct bpf_core_cand *new_cands, *cand;
> > - const struct btf_type *t;
> > - const char *targ_name;
> > + const struct btf_type *t, *local_t;
> > + const char *targ_name, *local_name;
> > size_t targ_essent_len;
> > int n, i;
> >
> > + local_t = btf__type_by_id(local_cand->btf, local_cand->id);
> > + local_name = btf__str_by_offset(local_cand->btf, local_t->name_off);
> > +
> > n = btf__type_cnt(targ_btf);
> > for (i = targ_start_id; i < n; i++) {
> > t = btf__type_by_id(targ_btf, i);
> > - if (btf_kind(t) != btf_kind(local_cand->t))
> > + if (btf_kind(t) != btf_kind(local_t))
> > continue;
> >
> > targ_name = btf__name_by_offset(targ_btf, t->name_off);
> > @@ -5198,12 +5201,12 @@ static int bpf_core_add_cands(struct
> > bpf_core_cand *local_cand,
> > if (targ_essent_len != local_essent_len)
> > continue;
> >
> > - if (strncmp(local_cand->name, targ_name, local_essent_len) != 0)
> > + if (strncmp(local_name, targ_name, local_essent_len) != 0)
> > continue;
> >
> > pr_debug("CO-RE relocating [%d] %s %s: found target candidate
> > [%d] %s %s in [%s]\n",
> > - local_cand->id, btf_kind_str(local_cand->t),
> > - local_cand->name, i, btf_kind_str(t), targ_name,
> > + local_cand->id, btf_kind_str(local_t),
> > + local_name, i, btf_kind_str(t), targ_name,
> > targ_btf_name);
> > new_cands = libbpf_reallocarray(cands->cands, cands->len + 1,
> > sizeof(*cands->cands));
> > @@ -5212,8 +5215,6 @@ static int bpf_core_add_cands(struct
> > bpf_core_cand *local_cand,
> >
> > cand = &new_cands[cands->len];
> > cand->btf = targ_btf;
> > - cand->t = t;
> > - cand->name = targ_name;
> > cand->id = i;
> >
> > cands->cands = new_cands;
> > @@ -5320,18 +5321,20 @@ bpf_core_find_cands(struct bpf_object *obj,
> > const struct btf *local_btf, __u32 l
> > struct bpf_core_cand local_cand = {};
> > struct bpf_core_cand_list *cands;
> > const struct btf *main_btf;
> > + const struct btf_type *local_t;
> > + const char *local_name;
> > size_t local_essent_len;
> > int err, i;
> >
> > local_cand.btf = local_btf;
> > - local_cand.t = btf__type_by_id(local_btf, local_type_id);
> > - if (!local_cand.t)
> > + local_t = btf__type_by_id(local_btf, local_type_id);
> > + if (!local_t)
> > return ERR_PTR(-EINVAL);
>
> Heh. Looks like you only compile-tested it :)
of course :)
> I was surprised that CO-RE in the kernel was working,
> but libbpf CO-RE didn't :)
> Thankfully the fix was simple:
>
> diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c
> index 24d1cbc30084..1341ce539662 100644
> --- a/tools/lib/bpf/libbpf.c
> +++ b/tools/lib/bpf/libbpf.c
> @@ -5333,6 +5333,7 @@ bpf_core_find_cands(struct bpf_object *obj,
> const struct btf *local_btf, __u32 l
> int err, i;
>
> local_cand.btf = local_btf;
> + local_cand.id = local_type_id;
fascinating, we didn't really use id because we just had type
directly, nice find, sorry about that
> local_t = btf__type_by_id(local_btf, local_type_id);
>
> Just fyi.
