On Fri, Feb 07, 2025 at 09:04:28PM +0100, Eric Woudstra wrote:

Or should mlxsw_sp_switchdev_blocking_event() use
switchdev_handle_port_obj_add_foreign() to add the vxlan
foreign port?

Then all foreign ports are added in a uniform manner and
SWITCHDEV_F_NO_FOREIGN is respected.

I do not have the hardware to test any changes in that code.

Personally, in your place I wouldn't have the courage to refactor that
much in a driver as complex as spectrum, but if you CC the right people
from Nvidia who can test, I guess you could give that a try.

Actually, how I came to spectrum was that I was thinking about an
alternative mechanism of detecting "foreign or not", other than emitting
two switchdev notifiers. You emit just the usual, single one, but
whoever handles it for a foreign bridge port will set a new bool
port_obj_info->handled_by_foreign, very similar to the existing
bool port_obj_info->handled. I was looking around to see who else
open-codes the switchdev object handling rather than use the
switchdev_handle_*() helpers, and that's how I came across spectrum.
It would seem, at first glance, easier to set just this in spectrum:

diff --git a/drivers/net/ethernet/mellanox/mlxsw/spectrum_switchdev.c b/drivers/net/ethernet/mellanox/mlxsw/spectrum_switchdev.c
index 6397ff0dc951..6926aaae7278 100644
--- a/drivers/net/ethernet/mellanox/mlxsw/spectrum_switchdev.c
+++ b/drivers/net/ethernet/mellanox/mlxsw/spectrum_switchdev.c

@@ -3953,6 +3953,7 @@ mlxsw_sp_switchdev_vxlan_vlans_add(struct net_device *vxlan_dev,
 		return 0;
 
 	port_obj_info->handled = true;
+	port_obj_info->handled_by_foreign = true;
 
 	bridge_device = mlxsw_sp_bridge_device_find(mlxsw_sp->bridge, br_dev);
 	if (!bridge_device)

and this in the object replication helper:

diff --git a/net/switchdev/switchdev.c b/net/switchdev/switchdev.c
index c48f66643e99..be82e79b5feb 100644
--- a/net/switchdev/switchdev.c
+++ b/net/switchdev/switchdev.c

@@ -763,6 +763,8 @@ static int __switchdev_handle_port_obj_add(struct net_device *dev,
 	if (!foreign_dev_check_cb(switchdev, dev))
 		return err;
 
+	port_obj_info->handled_by_foreign = true;
+
 	return __switchdev_handle_port_obj_add(br, port_obj_info, check_cb,
 					       foreign_dev_check_cb, add_cb);
 }

Just some care needs to be taken to only consider "handled_by_foreign"
just when "handled" is true.

I haven't yet decided which variant I like better, just thought I'd
mention this as something which requires a single switchdev notification.

Anyway, in the future I'll have to do some more tweaks with these flags
in the context of LAG. These flags (BR_VLFLAG_ADDED_BY_SWITCHDEV, now
also BR_VLFLAG_TAGGING_BY_SWITCHDEV after this patch) can dynamically
change, and the existing code isn't great because it doesn't handle that.

For example:

ip link add br0 type bridge
ip link set swp0 master br0
ip link set bond0 master br0 # bond0 is a foreign interface to swp0 at this time
bridge vlan add dev bond0 vid 100 # this won't get BR_VLFLAG_TAGGING_BY_SWITCHDEV
ip link set swp1 master bond0 # bond0 is no longer a foreign interface to swp0, assuming the same phys_switch_id
# vid 100 should get BR_VLFLAG_TAGGING_BY_SWITCHDEV during br_switchdev_vlan_replay()

Considering that br_switchdev_vlan_replay() will need to re-evaluate the
BR_VLFLAG_TAGGING_BY_SWITCHDEV flag, I guess I do prefer the simpler
variant after all - it is one call less that will have to be made during
replay as well.