One thought on the v3 delta that I missed earlier:

On Thu, Jan 24, 2019 at 01:15:18PM -0800, Suren Baghdasaryan wrote:

+/*
+ * psi_update_work represents slowpath accounting part while psi_group_change
+ * represents hotpath part. There are two potential races between them:
+ * 1. Changes to group->polling when slowpath checks for new stall, then hotpath
+ *    records new stall and then slowpath resets group->polling flag. This leads
+ *    to the exit from the polling mode while monitored state is still changing.
+ * 2. Slowpath overwriting an immediate update scheduled from the hotpath with
+ *    a regular update further in the future and missing the immediate update.
+ * Both races are handled with a retry cycle in the slowpath:
+ *
+ *    HOTPATH:                         |    SLOWPATH:
+ *                                     |
+ * A) times[cpu] += delta              | E) delta = times[*]
+ * B) start_poll = (delta[poll_mask] &&|    if delta[poll_mask]:
+ *      cmpxchg(g->polling, 0, 1) == 0)| F)   polling_until = now + grace_period
+ *    if start_poll:                   |    if now > polling_until:
+ * C)   mod_delayed_work(1)            |      if g->polling:

With the polling flag being atomic now, this "if g->polling" line
isn't accurate anymore. Since this diagram is specifically about
memory ordering, this should move the g->polling load up to where
delta is read and then refer to unordered local variables down here.