Re: [PATCH bpf-next v4 1/6] netlink: specs: Add XDP RX checksum capability to XDP metadata specs

[Stanislav Fomichev <sdf.kernel@gmail.com>]

On 07/14, Lorenzo Bianconi wrote:
> > On 07/08, Vladimir Vdovin wrote:
> > > From: Lorenzo Bianconi <lorenzo@kernel.org>
> > > 
> > > Introduce XDP RX checksum capability to XDP metadata specs. XDP RX
> > > checksum will be use by devices capable of exposing receive checksum
> > > result via bpf_xdp_metadata_rx_checksum().
> > > Moreover, introduce xmo_rx_checksum netdev callback in order to allow
> > > the eBPF program bound to the device to retrieve the RX checksum result
> > > computed by the hw NIC and reported via DMA descriptors.
> > > 
> > > Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
> > > Signed-off-by: Vladimir Vdovin <deliran@verdict.gg>
> > > ---
> > >  Documentation/netlink/specs/netdev.yaml |  5 ++++
> > >  include/net/xdp.h                       | 18 ++++++++++++++
> > >  include/uapi/linux/netdev.h             |  3 +++
> > >  net/core/xdp.c                          | 32 +++++++++++++++++++++++++
> > >  tools/include/uapi/linux/netdev.h       |  3 +++
> > >  5 files changed, 61 insertions(+)
> > > 
> > > diff --git a/Documentation/netlink/specs/netdev.yaml b/Documentation/netlink/specs/netdev.yaml
> > > index 5f143da7458c..6d0d90d3a614 100644
> > > --- a/Documentation/netlink/specs/netdev.yaml
> > > +++ b/Documentation/netlink/specs/netdev.yaml
> > > @@ -61,6 +61,11 @@ definitions:
> > >          doc: |
> > >            Device is capable of exposing receive packet VLAN tag via
> > >            bpf_xdp_metadata_rx_vlan_tag().
> > > +      -
> > > +        name: checksum
> > > +        doc: |
> > > +          Device is capable of exposing receive checksum result via
> > > +          bpf_xdp_metadata_rx_checksum().
> > >    -
> > >      type: flags
> > >      name: xsk-flags
> > > diff --git a/include/net/xdp.h b/include/net/xdp.h
> > > index aa742f413c35..e255ff786131 100644
> > > --- a/include/net/xdp.h
> > > +++ b/include/net/xdp.h
> > > @@ -586,6 +586,10 @@ void xdp_attachment_setup(struct xdp_attachment_info *info,
> > >  			   NETDEV_XDP_RX_METADATA_VLAN_TAG, \
> > >  			   bpf_xdp_metadata_rx_vlan_tag, \
> > >  			   xmo_rx_vlan_tag) \
> > > +	XDP_METADATA_KFUNC(XDP_METADATA_KFUNC_RX_CHECKSUM, \
> > > +			   NETDEV_XDP_RX_METADATA_CHECKSUM, \
> > > +			   bpf_xdp_metadata_rx_checksum, \
> > > +			   xmo_rx_checksum)
> > >  
> > >  enum xdp_rx_metadata {
> > >  #define XDP_METADATA_KFUNC(name, _, __, ___) name,
> > > @@ -643,12 +647,26 @@ enum xdp_rss_hash_type {
> > >  	XDP_RSS_TYPE_L4_IPV6_SCTP_EX = XDP_RSS_TYPE_L4_IPV6_SCTP | XDP_RSS_L3_DYNHDR,
> > >  };
> > 
> > [..]
> >  
> > > +/* Please note the driver is required to invalidate the checksum if the NIC
> > > + * reports CHECKSUM_UNNECESSARY or CHECKSUM_COMPLETE and the eBPF program
> > > + * modifies the packet since it can change some fields validated by the
> > > + * checksum.
> > > + */
> > 
> > Sorry, a bit confused about this part. IIUC this is in response to this
> > discussion https://lore.kernel.org/bpf/aaLYfWnuuf_ne72u@lore-desk/ ?
> > 
> > Looking at a few drivers:
> > - bnxt (bnxt_rx_pkt) does UNNECESSARY - ok
> > - mlx5 (mlx5e_handle_csum) does UNNECESSARY and skips COMPLETE if there is
> >   bpf prog attached
> > - fbnic (fbnic_rx_csum) - can do COMPLETE even with xdp attached?
> > - gve (gve_rx) - can do COMPLETE even with xdp attached?
> > 
> > So none of the drivers invalidate anything (and nether the bpf machinery), some
> > choose to (correctly?) not report COMPLETE when there is XDP. So was the
> > discussion about documenting this and giving some guidance? This probably
> > belongs to that `DOC: skb checksums` section of include/linux/skbuff.h?
> 
> Hi Stanislav,
> 
> re-reading the previous upstream discussion, I guess we should just document the
> expected driver behaviour (invalidate the cksum for both COMPLETE and
> UNNECESSARY cases when the XDP program running on the NIC returns XDP_PASS) and
> add a given kselftest for it. Agree?

Yes, let's have something a bit more detailed about how it works right now
vs the expectation.

By my point is: the comment in this patch also seems wrong? We do not seem
to invalidate the csum, we just disable COMPLETE when there is a xdp program
attached (or I'm missing something).