Re: [PATCH wireless-next v4 14/21] wifi: mac80211: add driver_op for SMD substate changes

From: Johannes Berg

Date: Mon Nov 16 2026 - 06:36:32 EST


On Sat, 2026-11-14 at 00:50 +0530, Pooventhiran G wrote:
>
> > > +void sta_info_reset_smd_state(struct sta_info *sta, bool notify)
> > > +{
> > > + if (sta->smd_state == IEEE80211_STA_SMD_NONE)
> > > + return;
> > > +
> > > + /* DRAIN -> NONE, EXEC -> NONE or PREP -> NONE */
> > > + WARN_ON_ONCE(sta_info_move_smd_state(sta, IEEE80211_STA_SMD_NONE,
> > > + notify));
> >
> > How is that even useful? If we're talking about a sub-state while in
> > state 4, then surely moving out of state 4 simply invalidates that, why
> > bother?
> >
>
> True, thought I could clean up the roam states before that exact invalidation
> happens. I will remove it from state 4->3 transition.
>
> > > + clear_sta_flag(sta, WLAN_STA_SMD_PREP_TARGET);
> > > + clear_sta_flag(sta, WLAN_STA_SMD_EXEC_CURRENT);
> > > + clear_sta_flag(sta, WLAN_STA_SMD_DL_DRAIN);
> >
> > Understand that you need to clear the stale flags, but for the driver
> > that just seems strange?
> >
>
> So, this is also called from state 4->3 transition so clearing the stale flags
> seems necessary. For the drv_sta_smd_state op as well, if we don't clear the
> state in mac80211, there will be inconsistency between driver and mac80211.

Wait, which way is it then? Anyway I'm not sure I see a lot of value in
moving the driver down when the whole state 4 is left later?

> > > + case IEEE80211_STA_SMD_EXEC:
> > > + case IEEE80211_STA_SMD_DRAIN:
> > > + err = drv_sta_smd_state(local, sta->sdata, &sta->sta,
> > > + IEEE80211_STA_SMD_NONE,
> > > + IEEE80211_STA_SMD_EXEC);
> > > + if (err)
> > > + goto rollback;
> > > +
> > > + if (new_ss == IEEE80211_STA_SMD_DRAIN)
> > > + err = drv_sta_smd_state(local, sta->sdata, &sta->sta,
> > > + IEEE80211_STA_SMD_EXEC,
> > > + IEEE80211_STA_SMD_DRAIN);
> > > + break;
> >
> > you do realize "fallthrough" exists?
> >
>
> I am afraid I am not able to understand how fallthrough helps here. 

My bad, sorry! I mixed up the order of operations.

> > > + /* if EXEC -> DRAIN failed in the driver, notify EXEC -> NONE */
> > > + WARN_ON_ONCE(drv_sta_smd_state(local, sta->sdata, &sta->sta,
> > > + IEEE80211_STA_SMD_EXEC,
> > > + IEEE80211_STA_SMD_NONE));
> >
> > I'm not a fan of sticking whole blocks of code into a WARN, but I'm also
> > not sure this whole "model what should be a one-way sub-state based on
> > sta_state" was a good idea in this patch.
> >
> Could you please kindly clarify "model what should be a one-way sub-state
> based on sta_state"? Would you prefer if the driver was able to directly
> update the state as well?

Ah, no, I was just mostly thinking about the above discussion, and also
the whole substates.

For station _state_ I think it makes sense (mostly) to go have the
"ladder" of states (notexist <-> none <-> auth <-> assoc <-> authorized)
and clearly modelled it based on that.

But does that makes sense here? Why even go 4 -> 4b -> 4c as a substate
etc. Wouldn't it be a lot simpler to just say "this is the new substate"
deal with it? There's not a lot of "ladder" going on, the substates
don't really depend on each other as fundamentally. In fact, for 4a vs.
b/c they're not even a ladder anyway since that's completely different
use cases.

johannes