On Mon, Jan 30, 2017 at 8:49 AM, Yotam Gigi <yot...@mellanox.com> wrote: > > Cong, after some thinking I think I don't really need the tcf_lock here. I > don't really care if the truncate, trunc_size, rate and tcf_action are > consistent among themselves - the only parameter that I care about is the > psample_group pointer, and it is protected via RCU. As far as I see, there is > no need to lock here.
OK, I trust you, you should know the logic better than me. > > I do need to take the tcf_lock to protect the statistics update in the > tcf_sample_act code, as far as I see. > Hm? You use percpu stats, so you don't need spinlock.
