On Fri, 2014-06-06 at 10:12 +0200, Michal Kazior wrote:
> On 6 June 2014 10:04, Coelho, Luciano <luciano.coelho@xxxxxxxxx> wrote:
> > On Fri, 2014-06-06 at 09:56 +0200, Michal Kazior wrote:
> >> On 6 June 2014 09:37, Coelho, Luciano <luciano.coelho@xxxxxxxxx> wrote:
> [...]
> >> > static u32 ieee80211_vif_queues_to_wake(struct ieee80211_local *local,
> >> > struct ieee80211_sub_if_data *sdata,
> >> > enum queue_stop_reason reason)
> >> > {
> >> > struct ieee80211_sub_if_data *sdata_iter;
> >> > unsigned long queues, queues_iter, q;
> >> > int i;
> >> >
> >> > lockdep_assert_held(&local->mtx);
> >> >
> >> > queues = ieee80211_get_vif_queues(local, sdata);
> >> >
> >> > rcu_read_lock();
> >> > list_for_each_entry_rcu(sdata_iter, &local->interfaces, list) {
> >> > if (!ieee80211_sdata_running(sdata_iter))
> >> > continue;
>
> Hmm, actually, shouldn't we `if (sdata == sdata_iter) continue`?
> Otherwise you'll never wake queues, will you?
Ahmmm... Yes of course. :)
> >> > queues_iter = ieee80211_get_vif_queues(local, sdata_iter);
> >> > for_each_set_bit(i, &queues_iter, local->hw.queues) {
> >> > q = sdata->vif.hw_queue[i];
> >> >
> >> > if (test_bit(reason, &local->queue_stop_reasons[q]))
> >> > queues &= ~q;
> >> > }
> >> > }
> >> > rcu_read_unlock();
> >> >
> >> > return queues;
> >> > }
> >> >
> >> > void ieee80211_wake_vif_queues(struct ieee80211_local *local,
> >> > struct ieee80211_sub_if_data *sdata,
> >> > enum queue_stop_reason reason)
> >> > {
> >> > unsigned int queues = ieee80211_vif_queues_to_wake(local, sdata, reason);
> >> >
> >> > ieee80211_wake_queues_by_reason(&local->hw, queues, reason);
> >> > }
> >>
> >> Once you set local->queue_stop_reasons[] for an overlapping queue you
> >> won't be able to wake it with ieee80211_wake_vif_queues() alone.
> >> You'll need to call ieee80211_wake_queues() to do that. Is that what
> >> you intended?
> >
> > More or less. My idea is that only the *last* vif that tries to wake
> > the overlapping queue, will actually do so.
>
> So ieee80211_wake_vif_queues() would be called regardless of the
> ieee80211_csa_needs_block_tx() and ieee80211_wake_queues() is called
> like it was before (i.e. when all block_tx interfaces are done).
Yes, ieee80211_wake_vif_queues() would be called regardless of the
ieee80211_csa_needs_block_tx(), but we should probably still check if
the vif in question has block_tx set?
And I don't see why we would have to call ieee80211_wake_queues() again
at the end... would that still be necessary?
--
Luca.
��.n��������+%������w��{.n�����{���zW����ܨ}���Ơz�j:+v�����w����ޙ��&�)ߡ�a����z�ޗ���ݢj��w�f
