[Intel-wired-lan] [PATCH net-next 08/12] ice: Add VLAN FDB support in switchdev mode

Drewek, Wojciech wojciech.drewek at intel.com
Thu Apr 27 10:28:21 UTC 2023



> -----Original Message-----
> From: Lobakin, Aleksander <aleksander.lobakin at intel.com>
> Sent: piątek, 21 kwietnia 2023 17:25
> To: Drewek, Wojciech <wojciech.drewek at intel.com>
> Cc: intel-wired-lan at lists.osuosl.org; netdev at vger.kernel.org; Lobakin, Aleksander <aleksander.lobakin at intel.com>; Ertman, David M
> <david.m.ertman at intel.com>; michal.swiatkowski at linux.intel.com; marcin.szycik at linux.intel.com; Chmielewski, Pawel
> <pawel.chmielewski at intel.com>; Samudrala, Sridhar <sridhar.samudrala at intel.com>
> Subject: Re: [PATCH net-next 08/12] ice: Add VLAN FDB support in switchdev mode
> 
> From: Wojciech Drewek <wojciech.drewek at intel.com>
> Date: Mon, 17 Apr 2023 11:34:08 +0200
> 
> > From: Marcin Szycik <marcin.szycik at intel.com>
> >
> > Add support for matching on VLAN tag in bridge offloads.
> > Currently only trunk mode is supported.
> >
> > To enable VLAN filtering (existing FDB entries will be deleted):
> > ip link set $BR type bridge vlan_filtering 1
> >
> > To add VLANs to bridge in trunk mode:
> > bridge vlan add dev $PF1 vid 110-111
> > bridge vlan add dev $VF1_PR vid 110-111
> >
> > Signed-off-by: Marcin Szycik <marcin.szycik at intel.com>
> > ---
> >  .../net/ethernet/intel/ice/ice_eswitch_br.c   | 319 +++++++++++++++++-
> >  .../net/ethernet/intel/ice/ice_eswitch_br.h   |  12 +
> >  2 files changed, 317 insertions(+), 14 deletions(-)
> >
> > diff --git a/drivers/net/ethernet/intel/ice/ice_eswitch_br.c b/drivers/net/ethernet/intel/ice/ice_eswitch_br.c
> > index 49381e4bf62a..56d36e397b12 100644
> > --- a/drivers/net/ethernet/intel/ice/ice_eswitch_br.c
> > +++ b/drivers/net/ethernet/intel/ice/ice_eswitch_br.c
> > @@ -59,13 +59,19 @@ ice_eswitch_br_netdev_to_port(struct net_device *dev)
> >  static void
> >  ice_eswitch_br_ingress_rule_setup(struct ice_adv_lkup_elem *list,
> >  				  struct ice_adv_rule_info *rule_info,
> > -				  const unsigned char *mac,
> > +				  const unsigned char *mac, bool vlan, u16 vid,
> 
> Could we use one combined argument? Doesn't `!!vid == !!vlan`? VID 0 is
> reserved IIRC...
> 
> (same in all the places below)

Makes sense

> 
> >  				  u8 pf_id, u16 vf_vsi_idx)
> >  {
> >  	list[0].type = ICE_MAC_OFOS;
> >  	ether_addr_copy(list[0].h_u.eth_hdr.dst_addr, mac);
> >  	eth_broadcast_addr(list[0].m_u.eth_hdr.dst_addr);
> 
> [...]
> 
> > @@ -344,10 +389,33 @@ ice_eswitch_br_fdb_entry_create(struct net_device *netdev,
> >  	struct device *dev = ice_pf_to_dev(pf);
> >  	struct ice_esw_br_fdb_entry *fdb_entry;
> >  	struct ice_esw_br_flow *flow;
> > +	struct ice_esw_br_vlan *vlan;
> >  	struct ice_hw *hw = &pf->hw;
> > +	bool add_vlan = false;
> >  	unsigned long event;
> >  	int err;
> >
> > +	/* FIXME: untagged filtering is not yet supported
> > +	 */
> 
> Shouldn't be present in release code I believe. I mean, the sentence is
> fine (just don't forget dot at the end), but without "FIXME:". And it
> can be one-liner.

Sure

> 
> > +	if (!(bridge->flags & ICE_ESWITCH_BR_VLAN_FILTERING) && vid)
> > +		return;
> 
> [...]
> 
> > +static void
> > +ice_eswitch_br_vlan_filtering_set(struct ice_esw_br *bridge, bool enable)
> > +{
> > +	bool filtering = bridge->flags & ICE_ESWITCH_BR_VLAN_FILTERING;
> > +
> > +	if (filtering == enable)
> > +		return;
> 
> 	if (enable == !!(bridge->flags & ICE_ESWITCH_BR_VLAN_FILTERING))
> 
> ?

I like it

> 
> > +
> > +	ice_eswitch_br_fdb_flush(bridge);
> > +	if (enable)
> > +		bridge->flags |= ICE_ESWITCH_BR_VLAN_FILTERING;
> > +	else
> > +		bridge->flags &= ~ICE_ESWITCH_BR_VLAN_FILTERING;
> > +}
> 
> [...]
> 
> > +	port = xa_load(&bridge->ports, vsi_idx);
> > +	if (!port)
> > +		return -EINVAL;
> > +
> > +	vlan = xa_load(&port->vlans, vid);
> > +	if (vlan) {
> > +		if (vlan->flags == flags)
> > +			return 0;
> > +
> > +		ice_eswitch_br_vlan_cleanup(port, vlan);
> > +	}
> > +
> > +	vlan = ice_eswitch_br_vlan_create(vid, flags, port);
> > +	if (IS_ERR(vlan)) {
> > +		NL_SET_ERR_MSG_MOD(extack, "Failed to create VLAN entry");
> 
> FYI, there's NL_SET_ERR_MSG_FMT_MOD() landed recently (a couple releases
> back), which supports format strings. E.g. you could pass VID, VSI ID,
> flags etc. there to have more meaningful output (right in userspace).

Sure, I guess I can improve log msgs in other patches now.

> 
> > +		return PTR_ERR(vlan);
> > +	}
> > +
> > +	return 0;
> > +}
> 
> [...]
> 
> > +static int
> > +ice_eswitch_br_port_obj_add(struct net_device *netdev, const void *ctx,
> > +			    const struct switchdev_obj *obj,
> > +			    struct netlink_ext_ack *extack)
> > +{
> > +	struct ice_esw_br_port *br_port = ice_eswitch_br_netdev_to_port(netdev);
> > +	struct switchdev_obj_port_vlan *vlan;
> > +	int err;
> > +
> > +	if (!br_port)
> > +		return -EINVAL;
> > +
> > +	switch (obj->id) {
> > +	case SWITCHDEV_OBJ_ID_PORT_VLAN:
> > +		vlan = SWITCHDEV_OBJ_PORT_VLAN(obj);
> > +		err = ice_eswitch_br_port_vlan_add(br_port->bridge,
> > +						   br_port->vsi_idx, vlan->vid,
> > +						   vlan->flags, extack);
> 
> return right here? You have `default` in the switch block, so the
> compiler shouldn't complain if you remove it from the end of the func.

Sure

> 
> > +		break;
> > +	default:
> > +		return -EOPNOTSUPP;
> > +	}
> > +
> > +	return err;
> > +}
> > +
> > +static int
> > +ice_eswitch_br_port_obj_del(struct net_device *netdev, const void *ctx,
> > +			    const struct switchdev_obj *obj)
> > +{
> > +	struct ice_esw_br_port *br_port = ice_eswitch_br_netdev_to_port(netdev);
> > +	struct switchdev_obj_port_vlan *vlan;
> > +
> > +	if (!br_port)
> > +		return -EINVAL;
> > +
> > +	switch (obj->id) {
> > +	case SWITCHDEV_OBJ_ID_PORT_VLAN:
> > +		vlan = SWITCHDEV_OBJ_PORT_VLAN(obj);
> > +		ice_eswitch_br_port_vlan_del(br_port->bridge, br_port->vsi_idx,
> > +					     vlan->vid);
> 
> (same)
> 
> > +		break;
> > +	default:
> > +		return -EOPNOTSUPP;
> > +	}
> > +
> > +	return 0;
> > +}
> > +
> > +static int
> > +ice_eswitch_br_port_obj_attr_set(struct net_device *netdev, const void *ctx,
> > +				 const struct switchdev_attr *attr,
> > +				 struct netlink_ext_ack *extack)
> > +{
> > +	struct ice_esw_br_port *br_port = ice_eswitch_br_netdev_to_port(netdev);
> > +
> > +	if (!br_port)
> > +		return -EINVAL;
> > +
> > +	switch (attr->id) {
> > +	case SWITCHDEV_ATTR_ID_BRIDGE_VLAN_FILTERING:
> > +		ice_eswitch_br_vlan_filtering_set(br_port->bridge,
> > +						  attr->u.vlan_filtering);
> 
> (and here)
> 
> > +		break;
> > +	default:
> > +		return -EOPNOTSUPP;
> > +	}
> > +
> > +	return 0;
> > +}
> 
> [...]
> 
> > +	br_offloads->switchdev_blk.notifier_call =
> > +		ice_eswitch_br_event_blocking;
> 
> Oh, you have two usages of ->switchdev_blk here, so you can add an
> intermediate variable to avoid line breaking, which would also shorten
> the line below :D
> 
> 	nb = &br_offloads->switchdev_blk;
> 	nb->notifier_call = ice_eswitch_br_event_blocking;
> 	...

Hmmm, I feel like it is more readable right now. It's clear that we're registering
switchdev blocking notifier block (switchdev_blk). Introducing generic variable (nb)
might a bit ambiguous IMO. So if you have nothing against it I'd leave it as it is.

> 
> > +	err = register_switchdev_blocking_notifier(&br_offloads->switchdev_blk);
> > +	if (err) {
> > +		dev_err(dev,
> > +			"Failed to register bridge blocking switchdev notifier\n");
> > +		goto err_reg_switchdev_blk;
> > +	}
> > +
> >  	br_offloads->netdev_nb.notifier_call = ice_eswitch_br_port_event;
> >  	err = register_netdevice_notifier(&br_offloads->netdev_nb);
> 
> (here the same, but no line breaks, so up to you. You could reuse the
>  same variable or leave it as it is)

(same here)

> 
> >  	if (err) {
> 
> [...]
> 
> > diff --git a/drivers/net/ethernet/intel/ice/ice_eswitch_br.h b/drivers/net/ethernet/intel/ice/ice_eswitch_br.h
> > index 73ad81bad655..cf3e2615a62a 100644
> > --- a/drivers/net/ethernet/intel/ice/ice_eswitch_br.h
> > +++ b/drivers/net/ethernet/intel/ice/ice_eswitch_br.h
> > @@ -42,10 +42,16 @@ struct ice_esw_br_port {
> >  	enum ice_esw_br_port_type type;
> >  	struct ice_vsi *vsi;
> >  	u16 vsi_idx;
> > +	struct xarray vlans;
> 
> Hmm, I feel like you can make ::type u16 and then stack it with
> ::vsi_idx, so that you avoid a hole here.
> 
> > +};
> > +
> > +enum {
> > +	ICE_ESWITCH_BR_VLAN_FILTERING = BIT(0),
> >  };
> >
> >  struct ice_esw_br {
> >  	struct ice_esw_br_offloads *br_offloads;
> > +	int flags;
> 
> Unsigned types fit flags better I think?

Yup

> 
> >  	int ifindex;
> 
> (BTW, ifindex is also usually unsigned unless it's not an error)

It is defined as int in net_device, so I don’t know

> 
> >
> >  	struct xarray ports;
> [...]
> 
> Thanks,
> Olek


More information about the Intel-wired-lan mailing list