On Thu, Jul 4, 2024 at 1:00 AM Sebastian Andrzej Siewior
<bigeasy@xxxxxxxxxxxxx> wrote:
>
> On 2024-07-03 14:39:16 [-0700], Alexei Starovoitov wrote:
> > > diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
> > > index 25ea393cf084b..f45b03706e4e9 100644
> > > --- a/include/uapi/linux/bpf.h
> > > +++ b/include/uapi/linux/bpf.h
> > > @@ -6290,12 +6290,12 @@ struct bpf_tunnel_key {
> > > */
> > > struct bpf_xfrm_state {
> > > __u32 reqid;
> > > - __u32 spi; /* Stored in network byte order */
> > > + __be32 spi; /* Stored in network byte order */
> > > __u16 family;
> > > __u16 ext; /* Padding, future use. */
> > > union {
> > > - __u32 remote_ipv4; /* Stored in network byte order */
> > > - __u32 remote_ipv6[4]; /* Stored in network byte order */
> > > + __be32 remote_ipv4; /* Stored in network byte order */
> > > + __be32 remote_ipv6[4]; /* Stored in network byte order */
> > > };
> > > };
> >
> > I don't think we should be changing uapi because of sparse.
> > I would ignore the warnings.
>
> There are other struct member within this bpf.h which use __be32 so it
> is known to userland (in terms of the compiler won't complain about an
> unknown type due to missing include). The type is essentially the same
> since the __bitwise attribute is empty except for sparse (which defines
> __CHECKER_).
> Therefore I wouldn't say this changes the uapi in an incompatible way.
There are two remote_ipv[46] fields in uapi/bpf.h added in 2016 and 2018
with __u32. We're not going to change one them to __be32 now,
because you noticed a sparse warn on the _kernel_ side.
In general, sparse warn is never a reason to change uapi.
If you want to shut up sparse, adjust the kernel side only,
or better yet ignore it.
