On Wed, Sep 16, 2020 at 05:19:26PM -0700, David Miller wrote: > From: Vladimir Oltean <olte...@gmail.com> > Date: Tue, 15 Sep 2020 21:22:24 +0300 > > > This is a problem because, at least theoretically, another timestampable > > skb might use the same ocelot_port->ts_id before that is incremented. So > > the logic of using and incrementing the timestamp id should be atomic > > per port. > > Have you actually observed this race in practice? > > All transmit calls are serialized by the netdev transmit spinlock. > > Let's not add locking if it is not actually necessary.
It's a bit more complicated. This code is also used from DSA, and DSA now declares NETIF_F_LLTX.
