On tis, maj 31, 2022 at 17:23, Ido Schimmel idosch@nvidia.com wrote:
On Tue, May 31, 2022 at 11:34:21AM +0200, Hans Schultz wrote:
Just to give you another data point about how this works in other devices, I can say that at least in Spectrum this works a bit differently. Packets that ingress via a locked port and incur an FDB miss are trapped to the CPU where they should be injected into the Rx path so that the bridge will create the 'locked' FDB entry and notify it to user space. The packets are obviously rated limited as the CPU cannot handle billions of packets per second, unlike the ASIC. The limit is not per bridge port (or even per bridge), but instead global to the entire device.
Btw, will the bridge not create a SWITCHDEV_FDB_ADD_TO_DEVICE event towards the switchcore in the scheme you mention and thus add an entry that opens up for the specified mac address?
It will, but the driver needs to ignore FDB entries that are notified with locked flag. I see that you extended 'struct switchdev_notifier_fdb_info' with the locked flag, but it's not initialized in br_switchdev_fdb_populate(). Can you add it in the next version?
An issue with sending the flag to the driver is that port_fdb_add() is suddenly getting more and more arguments and getting messy in my opinion, but maybe that's just how it is...
Another issue is that bridge fdb add MAC dev DEV master static seems to add the entry with the SELF flag set, which I don't think is what we would want it to do or? Also the replace command is not really supported properly as it is. I have made a fix for that which looks something like this:
diff --git a/net/bridge/br_fdb.c b/net/bridge/br_fdb.c index 6cbb27e3b976..f43aa204f375 100644 --- a/net/bridge/br_fdb.c +++ b/net/bridge/br_fdb.c @@ -917,6 +917,9 @@ static int fdb_add_entry(struct net_bridge *br, struct net_bridge_port *source, if (flags & NLM_F_EXCL) return -EEXIST;
+ if (flags & NLM_F_REPLACE) + modified = true; + if (READ_ONCE(fdb->dst) != source) { WRITE_ONCE(fdb->dst, source); modified = true;
The argument for always sending notifications to the driver in the case of replace is that a replace command will refresh the entries timeout if the entry is the same. Any thoughts on this?