Re: [Make-wifi-fast] [PATCH 1/2] ath9k: use mac80211 intermediate software queues

Re: [Make-wifi-fast] [PATCH 1/2] ath9k: use mac80211 intermediate software queues

[Toke Høiland-Jørgensen]

Tim Shepard <shep@alum.mit.edu> writes:

> Hmm... if the renaming is going to go in mainline, I feel pretty
> strongly it should go in *before* a patch to switch over to use the
> intermediate queues.  The whole point of the renaming was to make the
> code that uses the intermediate queues much more understandable
> (avoiding the unfortuante collision of "txq" meaning two different
> things throughout the code).
>
> Once it is all done and everyone's done reading and trying to
> understand this code, there's much less reason to do the renaming.
>
> Toke, how do you feel about this at this point?

I'm fine with not renaming things for now. Been looking at the current
code enough that it doesn't bother me.

Oh, and you can hide most of the ieee80211_txq stuff behind macros, so
it doesn't have to be all over the code. Makes the patch set smaller
too...

> I'm asking because I hope to have a new version of my patch soon
> (fixing a bug in how it handles tid->hwq->pending_frames and
> hq_max_pending[*] ),

Cool. I started looking into what it will take to do a full conversion
(getting rid of the old TX path). Not quite there yet (to say the
least), so if you have a less buggy base I can work from that would be
cool ;)

-Toke

Re: [Make-wifi-fast] [PATCH 1/2] ath9k: use mac80211 intermediate software queues

[Felix Fietkau]

On 2016-06-17 15:41, Tim Shepard wrote:
>> > diff --git a/drivers/net/wireless/ath/ath9k/ath9k.h b/drivers/net/wireless/ath/ath9k/ath9k.h
>> > index 93b3793..caeae10 100644
>> > --- a/drivers/net/wireless/ath/ath9k/ath9k.h
>> > +++ b/drivers/net/wireless/ath/ath9k/ath9k.h
>> > @@ -145,8 +145,6 @@ int ath_descdma_setup(struct ath_softc *sc, struct ath_descdma *dd,
>> >  #define BAW_WITHIN(_start, _bawsz, _seqno) \
>> >  	((((_seqno) - (_start)) & 4095) < (_bawsz))
>> >  
>> > -#define ATH_AN_2_TID(_an, _tidno)  (&(_an)->tid[(_tidno)])
>> > -
>> >  #define IS_HT_RATE(rate)   (rate & 0x80)
>> >  #define IS_CCK_RATE(rate)  ((rate >= 0x18) && (rate <= 0x1e))
>> >  #define IS_OFDM_RATE(rate) ((rate >= 0x8) && (rate <= 0xf))
>> > @@ -232,8 +230,10 @@ struct ath_buf {
>> >  
>> >  struct ath_atx_tid {
>> >  	struct list_head list;
>> > +	struct sk_buff_head i_q;
>> Do we really need a third queue here? Instead of adding yet another
>> layer of queueing here, I think we should even get rid of buf_q.
>> 
>> Channel context based queue handling can be dealt with by
>> stopping/starting relevant queues on channel context changes.
>> 
>> buf_q becomes unnecessary when you remove all code in the drv_tx
>> codepath that moves frames to the intermediate queue.
>> 
>> Any frame that was pulled from the intermediate queue and prepared for
>> tx, but which can't be sent right now can simply be queued to retry_q.
>> 
>> This will also help with getting the diffstat insertion/deletion ratio
>> under control ;)
>> 
>> >  	struct sk_buff_head buf_q;
>> >  	struct sk_buff_head retry_q;
>> > +	struct ieee80211_txq *swq;
>> No need for this pointer, you can use container_of.
> 
> 
> Felix, great to hear from you and thanks for your feedback.  I will
> try to work on this.
> 
> I was struggling to understand the channel context stuff, and I have
> no idea how to test it.  (Is there anyone else listening who might be
> able to help with testing the channel context stuff as we improve this
> patch and simplify the ath9k driver's use of the new mac80211
> intermediate queues?)
> 
> 
> Felix, do you have any thoughts on the renaming of txq to hwx that I
> had done in my original version of this patch?  I had a good e-mail
> discussion with Toke a week or two ago (cc these same various lists)
> and I believe he came to understand that perhaps the renaming I had
> done in the original version of this patch was worth doing.
> 
> Now in Toke's version of my patch he calls the ieee80211 txq a "swq"
> and the ath9k hardware queue is called a "txq".  (I had called the
> ieee80211 txq a "txq" and I renamed the ath9k hardware queue "hwq"
> throught all the ath9k driver code.    This also made ath9k's names of
> things more similar to mt76 which I was looking at as an example of a
> driver that uses your new ieee80211 txq mechanism.
> 
> I think the renaming is worth doing, but I also understand the
> renaming can be disruptive to others actively working on ath9k.
> It would be nice to have another opinion on this.
I think we should finish intermediate queues support first and then look
into the rename later.

- Felix

Re: [Make-wifi-fast] [PATCH 1/2] ath9k: use mac80211 intermediate software queues

[Felix Fietkau]

On 2016-06-17 11:09, Toke Høiland-Jørgensen wrote:
> This patch leaves the code for ath9k's internal per-node per-tid
> queues in place and just modifies the driver to also pull from
> the new mac80211 intermediate software queues, and implements
> the .wake_tx_queue method, which will cause mac80211 to deliver
> packets to be sent via the new intermediate queue.
> 
> Signed-off-by: Tim Shepard <shep@alum.mit.edu>
> 
> Reworked to not require the global variable renaming in ath9k.
> 
> Signed-off-by: Toke Høiland-Jørgensen <toke@toke.dk>
> ---
>  drivers/net/wireless/ath/ath9k/ath9k.h     |  16 +++-
>  drivers/net/wireless/ath/ath9k/debug_sta.c |   7 +-
>  drivers/net/wireless/ath/ath9k/init.c      |   1 +
>  drivers/net/wireless/ath/ath9k/main.c      |   1 +
>  drivers/net/wireless/ath/ath9k/xmit.c      | 119 +++++++++++++++++++++++++----
>  5 files changed, 125 insertions(+), 19 deletions(-)
> 
> diff --git a/drivers/net/wireless/ath/ath9k/ath9k.h b/drivers/net/wireless/ath/ath9k/ath9k.h
> index 93b3793..caeae10 100644
> --- a/drivers/net/wireless/ath/ath9k/ath9k.h
> +++ b/drivers/net/wireless/ath/ath9k/ath9k.h
> @@ -145,8 +145,6 @@ int ath_descdma_setup(struct ath_softc *sc, struct ath_descdma *dd,
>  #define BAW_WITHIN(_start, _bawsz, _seqno) \
>  	((((_seqno) - (_start)) & 4095) < (_bawsz))
>  
> -#define ATH_AN_2_TID(_an, _tidno)  (&(_an)->tid[(_tidno)])
> -
>  #define IS_HT_RATE(rate)   (rate & 0x80)
>  #define IS_CCK_RATE(rate)  ((rate >= 0x18) && (rate <= 0x1e))
>  #define IS_OFDM_RATE(rate) ((rate >= 0x8) && (rate <= 0xf))
> @@ -232,8 +230,10 @@ struct ath_buf {
>  
>  struct ath_atx_tid {
>  	struct list_head list;
> +	struct sk_buff_head i_q;
Do we really need a third queue here? Instead of adding yet another
layer of queueing here, I think we should even get rid of buf_q.

Channel context based queue handling can be dealt with by
stopping/starting relevant queues on channel context changes.

buf_q becomes unnecessary when you remove all code in the drv_tx
codepath that moves frames to the intermediate queue.

Any frame that was pulled from the intermediate queue and prepared for
tx, but which can't be sent right now can simply be queued to retry_q.

This will also help with getting the diffstat insertion/deletion ratio
under control ;)

>  	struct sk_buff_head buf_q;
>  	struct sk_buff_head retry_q;
> +	struct ieee80211_txq *swq;
No need for this pointer, you can use container_of.

- Felix

Re: [Make-wifi-fast] [PATCH 1/2] ath9k: use mac80211 intermediate software queues

[Toke Høiland-Jørgensen]

Felix Fietkau <nbd@nbd.name> writes:

> On 2016-06-17 11:09, Toke Høiland-Jørgensen wrote:
>> This patch leaves the code for ath9k's internal per-node per-tid
>> queues in place and just modifies the driver to also pull from
>> the new mac80211 intermediate software queues, and implements
>> the .wake_tx_queue method, which will cause mac80211 to deliver
>> packets to be sent via the new intermediate queue.
>> 
>> Signed-off-by: Tim Shepard <shep@alum.mit.edu>
>> 
>> Reworked to not require the global variable renaming in ath9k.
>> 
>> Signed-off-by: Toke Høiland-Jørgensen <toke@toke.dk>
>> ---
>>  drivers/net/wireless/ath/ath9k/ath9k.h     |  16 +++-
>>  drivers/net/wireless/ath/ath9k/debug_sta.c |   7 +-
>>  drivers/net/wireless/ath/ath9k/init.c      |   1 +
>>  drivers/net/wireless/ath/ath9k/main.c      |   1 +
>>  drivers/net/wireless/ath/ath9k/xmit.c      | 119 +++++++++++++++++++++++++----
>>  5 files changed, 125 insertions(+), 19 deletions(-)
>> 
>> diff --git a/drivers/net/wireless/ath/ath9k/ath9k.h b/drivers/net/wireless/ath/ath9k/ath9k.h
>> index 93b3793..caeae10 100644
>> --- a/drivers/net/wireless/ath/ath9k/ath9k.h
>> +++ b/drivers/net/wireless/ath/ath9k/ath9k.h
>> @@ -145,8 +145,6 @@ int ath_descdma_setup(struct ath_softc *sc, struct ath_descdma *dd,
>>  #define BAW_WITHIN(_start, _bawsz, _seqno) \
>>  	((((_seqno) - (_start)) & 4095) < (_bawsz))
>>  
>> -#define ATH_AN_2_TID(_an, _tidno)  (&(_an)->tid[(_tidno)])
>> -
>>  #define IS_HT_RATE(rate)   (rate & 0x80)
>>  #define IS_CCK_RATE(rate)  ((rate >= 0x18) && (rate <= 0x1e))
>>  #define IS_OFDM_RATE(rate) ((rate >= 0x8) && (rate <= 0xf))
>> @@ -232,8 +230,10 @@ struct ath_buf {
>>  
>>  struct ath_atx_tid {
>>  	struct list_head list;
>> +	struct sk_buff_head i_q;
> Do we really need a third queue here? Instead of adding yet another
> layer of queueing here, I think we should even get rid of buf_q.

This is definitely something that needs to be improved. One other
sticking point related to this: in the current version of this patch
ath_tid_has_buffered() gains a side effect of pulling from the mac80211
txq, which is obviously not so nice.

The obvious way to get rid of this is to export a txq_has_buffered()
function at the mac80211 layer. But avoiding that may be possible; the
sticking point is what to do with the code paths that do not dequeue
packets, but check ath_tid_has_buffered() to decide whether to schedule
the queue and/or to tell ieee80211_sta_set_buffered() about it (these
are for instance ath_tx_aggr_sleep/wakeup(). Can those just be removed
(i.e. don't call into ieee80211, and always schedule the txq on wakeup?
I'm not familiar enough with the intermediate queues to make that
call...


> Channel context based queue handling can be dealt with by
> stopping/starting relevant queues on channel context changes.

Noted.

> buf_q becomes unnecessary when you remove all code in the drv_tx
> codepath that moves frames to the intermediate queue.
>
> Any frame that was pulled from the intermediate queue and prepared for
> tx, but which can't be sent right now can simply be queued to retry_q.

Right.

> This will also help with getting the diffstat insertion/deletion ratio
> under control ;)

Yes, that would be good ;)

>>  	struct sk_buff_head buf_q;
>>  	struct sk_buff_head retry_q;
>> +	struct ieee80211_txq *swq;
> No need for this pointer, you can use container_of.

Ah, cool, thanks!

-Toke

Re: [Make-wifi-fast] [PATCH 1/2] ath9k: use mac80211 intermediate software queues

[Felix Fietkau]

On 2016-06-17 15:43, Toke Høiland-Jørgensen wrote:
> Felix Fietkau <nbd@nbd.name> writes:
> 
>> On 2016-06-17 11:09, Toke Høiland-Jørgensen wrote:
>>> This patch leaves the code for ath9k's internal per-node per-tid
>>> queues in place and just modifies the driver to also pull from
>>> the new mac80211 intermediate software queues, and implements
>>> the .wake_tx_queue method, which will cause mac80211 to deliver
>>> packets to be sent via the new intermediate queue.
>>> 
>>> Signed-off-by: Tim Shepard <shep@alum.mit.edu>
>>> 
>>> Reworked to not require the global variable renaming in ath9k.
>>> 
>>> Signed-off-by: Toke Høiland-Jørgensen <toke@toke.dk>
>>> ---
>>>  drivers/net/wireless/ath/ath9k/ath9k.h     |  16 +++-
>>>  drivers/net/wireless/ath/ath9k/debug_sta.c |   7 +-
>>>  drivers/net/wireless/ath/ath9k/init.c      |   1 +
>>>  drivers/net/wireless/ath/ath9k/main.c      |   1 +
>>>  drivers/net/wireless/ath/ath9k/xmit.c      | 119 +++++++++++++++++++++++++----
>>>  5 files changed, 125 insertions(+), 19 deletions(-)
>>> 
>>> diff --git a/drivers/net/wireless/ath/ath9k/ath9k.h b/drivers/net/wireless/ath/ath9k/ath9k.h
>>> index 93b3793..caeae10 100644
>>> --- a/drivers/net/wireless/ath/ath9k/ath9k.h
>>> +++ b/drivers/net/wireless/ath/ath9k/ath9k.h
>>> @@ -145,8 +145,6 @@ int ath_descdma_setup(struct ath_softc *sc, struct ath_descdma *dd,
>>>  #define BAW_WITHIN(_start, _bawsz, _seqno) \
>>>  	((((_seqno) - (_start)) & 4095) < (_bawsz))
>>>  
>>> -#define ATH_AN_2_TID(_an, _tidno)  (&(_an)->tid[(_tidno)])
>>> -
>>>  #define IS_HT_RATE(rate)   (rate & 0x80)
>>>  #define IS_CCK_RATE(rate)  ((rate >= 0x18) && (rate <= 0x1e))
>>>  #define IS_OFDM_RATE(rate) ((rate >= 0x8) && (rate <= 0xf))
>>> @@ -232,8 +230,10 @@ struct ath_buf {
>>>  
>>>  struct ath_atx_tid {
>>>  	struct list_head list;
>>> +	struct sk_buff_head i_q;
>> Do we really need a third queue here? Instead of adding yet another
>> layer of queueing here, I think we should even get rid of buf_q.
> 
> This is definitely something that needs to be improved. One other
> sticking point related to this: in the current version of this patch
> ath_tid_has_buffered() gains a side effect of pulling from the mac80211
> txq, which is obviously not so nice.
> 
> The obvious way to get rid of this is to export a txq_has_buffered()
> function at the mac80211 layer. But avoiding that may be possible; the
> sticking point is what to do with the code paths that do not dequeue
> packets, but check ath_tid_has_buffered() to decide whether to schedule
> the queue and/or to tell ieee80211_sta_set_buffered() about it (these
> are for instance ath_tx_aggr_sleep/wakeup(). Can those just be removed
> (i.e. don't call into ieee80211, and always schedule the txq on wakeup?
> I'm not familiar enough with the intermediate queues to make that
> call...
For tx scheduling, we can use swq_nonempty and deal with false positives.
For power save we should only use ieee80211_sta_set_buffered if the
driver itself has buffered some frames. Indication of packets in the
mac80211 intermediate queue is already taken care of inside mac80211.

- Felix

Re: [Make-wifi-fast] [PATCH 1/2] ath9k: use mac80211 intermediate software queues

[Felix Fietkau]

On 2016-06-17 15:48, Felix Fietkau wrote:
> On 2016-06-17 15:43, Toke Høiland-Jørgensen wrote:
>> Felix Fietkau <nbd@nbd.name> writes:
>> 
>>> On 2016-06-17 11:09, Toke Høiland-Jørgensen wrote:
>>>> This patch leaves the code for ath9k's internal per-node per-tid
>>>> queues in place and just modifies the driver to also pull from
>>>> the new mac80211 intermediate software queues, and implements
>>>> the .wake_tx_queue method, which will cause mac80211 to deliver
>>>> packets to be sent via the new intermediate queue.
>>>> 
>>>> Signed-off-by: Tim Shepard <shep@alum.mit.edu>
>>>> 
>>>> Reworked to not require the global variable renaming in ath9k.
>>>> 
>>>> Signed-off-by: Toke Høiland-Jørgensen <toke@toke.dk>
>>>> ---
>>>>  drivers/net/wireless/ath/ath9k/ath9k.h     |  16 +++-
>>>>  drivers/net/wireless/ath/ath9k/debug_sta.c |   7 +-
>>>>  drivers/net/wireless/ath/ath9k/init.c      |   1 +
>>>>  drivers/net/wireless/ath/ath9k/main.c      |   1 +
>>>>  drivers/net/wireless/ath/ath9k/xmit.c      | 119 +++++++++++++++++++++++++----
>>>>  5 files changed, 125 insertions(+), 19 deletions(-)
>>>> 
>>>> diff --git a/drivers/net/wireless/ath/ath9k/ath9k.h b/drivers/net/wireless/ath/ath9k/ath9k.h
>>>> index 93b3793..caeae10 100644
>>>> --- a/drivers/net/wireless/ath/ath9k/ath9k.h
>>>> +++ b/drivers/net/wireless/ath/ath9k/ath9k.h
>>>> @@ -145,8 +145,6 @@ int ath_descdma_setup(struct ath_softc *sc, struct ath_descdma *dd,
>>>>  #define BAW_WITHIN(_start, _bawsz, _seqno) \
>>>>  	((((_seqno) - (_start)) & 4095) < (_bawsz))
>>>>  
>>>> -#define ATH_AN_2_TID(_an, _tidno)  (&(_an)->tid[(_tidno)])
>>>> -
>>>>  #define IS_HT_RATE(rate)   (rate & 0x80)
>>>>  #define IS_CCK_RATE(rate)  ((rate >= 0x18) && (rate <= 0x1e))
>>>>  #define IS_OFDM_RATE(rate) ((rate >= 0x8) && (rate <= 0xf))
>>>> @@ -232,8 +230,10 @@ struct ath_buf {
>>>>  
>>>>  struct ath_atx_tid {
>>>>  	struct list_head list;
>>>> +	struct sk_buff_head i_q;
>>> Do we really need a third queue here? Instead of adding yet another
>>> layer of queueing here, I think we should even get rid of buf_q.
>> 
>> This is definitely something that needs to be improved. One other
>> sticking point related to this: in the current version of this patch
>> ath_tid_has_buffered() gains a side effect of pulling from the mac80211
>> txq, which is obviously not so nice.
>> 
>> The obvious way to get rid of this is to export a txq_has_buffered()
>> function at the mac80211 layer. But avoiding that may be possible; the
>> sticking point is what to do with the code paths that do not dequeue
>> packets, but check ath_tid_has_buffered() to decide whether to schedule
>> the queue and/or to tell ieee80211_sta_set_buffered() about it (these
>> are for instance ath_tx_aggr_sleep/wakeup(). Can those just be removed
>> (i.e. don't call into ieee80211, and always schedule the txq on wakeup?
>> I'm not familiar enough with the intermediate queues to make that
>> call...
> For tx scheduling, we can use swq_nonempty and deal with false positives.
> For power save we should only use ieee80211_sta_set_buffered if the
> driver itself has buffered some frames. Indication of packets in the
> mac80211 intermediate queue is already taken care of inside mac80211.
One more thing that I forgot in my previous reply: on PS wakeup, the
driver does not need to schedule the intermediate queues itself -
mac80211 will call drv_wake_tx_queue if frames are pending.

- Felix

[Make-wifi-fast] [PATCH 0/2] ath9k: Add airtime fairness scheduler

[Toke Høiland-Jørgensen]

(Re-send to make-wifi-fast list due to a busted MX record).

This is the second version of my airtime fairness patch. This version
has a somewhat reworked scheduler (now closer to the structure of
fq_codel) and a different way to measure RX airtime; and there's a
debugfs entry to control which airtime measurements to include in the
scheduling decisions. For a simple one-way UDP test, the scheduler
achieves pretty much perfect airtime share (by its own measure). There's
not much throughput difference in the UDP case, but TCP tests see a
moderate improvement. I'll write up something more detailed on the
performance measures over the weekend and post it in a separate mail.

This patch set is rebased to mac80211-next - which means it no longer
includes Michal's patch to disable qdiscs. I have retained my version of
Tim's patch to make ath9k use wake_tx_queue in this patch set. That
probably needs some work still, but I believe he is working on that. I
have not tested extensively with the mac80211 FQ-CoDel patches enabled,
but I expect them to be complementary to this.

Changes since the RFC version:

- The scheduler will now enforce fairness harder. The previous version
  would refill the deficit of slow stations too fast in some cases.

- Change the way RX airtime is measured. For aggregates, the airtime is
  now calculated as the difference between the rs->rs_tstamp of the
  first and last frame in the aggregate. For non-aggregates, the
  previous calculation from the packet size is retained.

- There is now an 'airtime_flags' debugfs entry which can be used to
  control which airtime measures are accounted to the deficit. If bit 0
  is set, TX airtime will be accounted, and if bit 1 is set, RX airtime
  will. If no bits are set, the scheduler will revert to simple
  round-robin scheduling. The default is enabling both TX and RX.

- Squashed the whole thing into one patch and rebased to mac80211-next.

Toke Høiland-Jørgensen (2):
  ath9k: use mac80211 intermediate software queues
  ath9k: Add a per-station airtime deficit scheduler

 drivers/net/wireless/ath/ath9k/ath9k.h     |  34 +++-
 drivers/net/wireless/ath/ath9k/channel.c   |  12 +-
 drivers/net/wireless/ath/ath9k/debug.c     |   3 +
 drivers/net/wireless/ath/ath9k/debug.h     |  29 ++++
 drivers/net/wireless/ath/ath9k/debug_sta.c |  53 +++++-
 drivers/net/wireless/ath/ath9k/init.c      |   2 +
 drivers/net/wireless/ath/ath9k/main.c      |   7 +-
 drivers/net/wireless/ath/ath9k/recv.c      |  60 +++++++
 drivers/net/wireless/ath/ath9k/xmit.c      | 255 ++++++++++++++++++++++-------
 9 files changed, 386 insertions(+), 69 deletions(-)

-- 
2.8.3

[Make-wifi-fast] [PATCH 1/2] ath9k: use mac80211 intermediate software queues

[Toke Høiland-Jørgensen]

This patch leaves the code for ath9k's internal per-node per-tid
queues in place and just modifies the driver to also pull from
the new mac80211 intermediate software queues, and implements
the .wake_tx_queue method, which will cause mac80211 to deliver
packets to be sent via the new intermediate queue.

Signed-off-by: Tim Shepard <shep@alum.mit.edu>

Reworked to not require the global variable renaming in ath9k.

Signed-off-by: Toke Høiland-Jørgensen <toke@toke.dk>
---
 drivers/net/wireless/ath/ath9k/ath9k.h     |  16 +++-
 drivers/net/wireless/ath/ath9k/debug_sta.c |   7 +-
 drivers/net/wireless/ath/ath9k/init.c      |   1 +
 drivers/net/wireless/ath/ath9k/main.c      |   1 +
 drivers/net/wireless/ath/ath9k/xmit.c      | 119 +++++++++++++++++++++++++----
 5 files changed, 125 insertions(+), 19 deletions(-)

diff --git a/drivers/net/wireless/ath/ath9k/ath9k.h b/drivers/net/wireless/ath/ath9k/ath9k.h
index 93b3793..caeae10 100644
--- a/drivers/net/wireless/ath/ath9k/ath9k.h
+++ b/drivers/net/wireless/ath/ath9k/ath9k.h
@@ -145,8 +145,6 @@ int ath_descdma_setup(struct ath_softc *sc, struct ath_descdma *dd,
 #define BAW_WITHIN(_start, _bawsz, _seqno) \
 	((((_seqno) - (_start)) & 4095) < (_bawsz))
 
-#define ATH_AN_2_TID(_an, _tidno)  (&(_an)->tid[(_tidno)])
-
 #define IS_HT_RATE(rate)   (rate & 0x80)
 #define IS_CCK_RATE(rate)  ((rate >= 0x18) && (rate <= 0x1e))
 #define IS_OFDM_RATE(rate) ((rate >= 0x8) && (rate <= 0xf))
@@ -232,8 +230,10 @@ struct ath_buf {
 
 struct ath_atx_tid {
 	struct list_head list;
+	struct sk_buff_head i_q;
 	struct sk_buff_head buf_q;
 	struct sk_buff_head retry_q;
+	struct ieee80211_txq *swq;
 	struct ath_node *an;
 	struct ath_txq *txq;
 	unsigned long tx_buf[BITS_TO_LONGS(ATH_TID_MAX_BUFS)];
@@ -247,13 +247,13 @@ struct ath_atx_tid {
 	s8 bar_index;
 	bool active;
 	bool clear_ps_filter;
+	bool swq_nonempty;
 };
 
 struct ath_node {
 	struct ath_softc *sc;
 	struct ieee80211_sta *sta; /* station struct we're part of */
 	struct ieee80211_vif *vif; /* interface with which we're associated */
-	struct ath_atx_tid tid[IEEE80211_NUM_TIDS];
 
 	u16 maxampdu;
 	u8 mpdudensity;
@@ -271,6 +271,15 @@ struct ath_node {
 	struct list_head list;
 };
 
+static inline
+struct ath_atx_tid *ath_an_2_tid(struct ath_node *an, u8 tidno)
+{
+	struct ieee80211_sta *sta = an->sta;
+	struct ieee80211_vif *vif = an->vif;
+	struct ieee80211_txq *swq = sta ? sta->txq[tidno] : vif->txq;
+	return (struct ath_atx_tid *) swq->drv_priv;
+}
+
 struct ath_tx_control {
 	struct ath_txq *txq;
 	struct ath_node *an;
@@ -585,6 +594,7 @@ void ath9k_release_buffered_frames(struct ieee80211_hw *hw,
 				   u16 tids, int nframes,
 				   enum ieee80211_frame_release_type reason,
 				   bool more_data);
+void ath9k_wake_tx_queue(struct ieee80211_hw *hw, struct ieee80211_txq *swq);
 
 /********/
 /* VIFs */
diff --git a/drivers/net/wireless/ath/ath9k/debug_sta.c b/drivers/net/wireless/ath/ath9k/debug_sta.c
index b66cfa9..0e7f6b5 100644
--- a/drivers/net/wireless/ath/ath9k/debug_sta.c
+++ b/drivers/net/wireless/ath/ath9k/debug_sta.c
@@ -25,6 +25,7 @@ static ssize_t read_file_node_aggr(struct file *file, char __user *user_buf,
 {
 	struct ath_node *an = file->private_data;
 	struct ath_softc *sc = an->sc;
+	struct ieee80211_txq *swq;
 	struct ath_atx_tid *tid;
 	struct ath_txq *txq;
 	u32 len = 0, size = 4096;
@@ -52,8 +53,10 @@ static ssize_t read_file_node_aggr(struct file *file, char __user *user_buf,
 			 "TID", "SEQ_START", "SEQ_NEXT", "BAW_SIZE",
 			 "BAW_HEAD", "BAW_TAIL", "BAR_IDX", "SCHED", "PAUSED");
 
-	for (tidno = 0, tid = &an->tid[tidno];
-	     tidno < IEEE80211_NUM_TIDS; tidno++, tid++) {
+	for (tidno = 0;
+	     tidno < IEEE80211_NUM_TIDS; tidno++) {
+		swq = an->sta->txq[tidno];
+		tid = (struct ath_atx_tid *) swq->drv_priv;
 		txq = tid->txq;
 		ath_txq_lock(sc, txq);
 		if (tid->active) {
diff --git a/drivers/net/wireless/ath/ath9k/init.c b/drivers/net/wireless/ath/ath9k/init.c
index 2ee8624..211736c 100644
--- a/drivers/net/wireless/ath/ath9k/init.c
+++ b/drivers/net/wireless/ath/ath9k/init.c
@@ -873,6 +873,7 @@ static void ath9k_set_hw_capab(struct ath_softc *sc, struct ieee80211_hw *hw)
 	hw->max_rate_tries = 10;
 	hw->sta_data_size = sizeof(struct ath_node);
 	hw->vif_data_size = sizeof(struct ath_vif);
+	hw->txq_data_size = sizeof(struct ath_atx_tid);
 	hw->extra_tx_headroom = 4;
 
 	hw->wiphy->available_antennas_rx = BIT(ah->caps.max_rxchains) - 1;
diff --git a/drivers/net/wireless/ath/ath9k/main.c b/drivers/net/wireless/ath/ath9k/main.c
index 8b63988..6ab56e5 100644
--- a/drivers/net/wireless/ath/ath9k/main.c
+++ b/drivers/net/wireless/ath/ath9k/main.c
@@ -2668,4 +2668,5 @@ struct ieee80211_ops ath9k_ops = {
 	.sw_scan_start	    = ath9k_sw_scan_start,
 	.sw_scan_complete   = ath9k_sw_scan_complete,
 	.get_txpower        = ath9k_get_txpower,
+	.wake_tx_queue      = ath9k_wake_tx_queue,
 };
diff --git a/drivers/net/wireless/ath/ath9k/xmit.c b/drivers/net/wireless/ath/ath9k/xmit.c
index 8ddd604..cdc8684 100644
--- a/drivers/net/wireless/ath/ath9k/xmit.c
+++ b/drivers/net/wireless/ath/ath9k/xmit.c
@@ -65,6 +65,8 @@ static struct ath_buf *ath_tx_setup_buffer(struct ath_softc *sc,
 					   struct ath_txq *txq,
 					   struct ath_atx_tid *tid,
 					   struct sk_buff *skb);
+static int ath_tx_prepare(struct ieee80211_hw *hw, struct sk_buff *skb,
+			  struct ath_tx_control *txctl);
 
 enum {
 	MCS_HT20,
@@ -118,6 +120,21 @@ static void ath_tx_queue_tid(struct ath_softc *sc, struct ath_txq *txq,
 		list_add_tail(&tid->list, list);
 }
 
+void ath9k_wake_tx_queue(struct ieee80211_hw *hw, struct ieee80211_txq *swq)
+{
+	struct ath_softc *sc = hw->priv;
+	struct ath_atx_tid *tid = (struct ath_atx_tid *) swq->drv_priv;
+	struct ath_txq *txq = tid->txq;
+
+	spin_lock_bh(&txq->axq_lock);
+
+	tid->swq_nonempty = true;
+	ath_tx_queue_tid(sc, txq, tid);
+	ath_txq_schedule(sc, txq);
+
+	spin_unlock_bh(&txq->axq_lock);
+}
+
 static struct ath_frame_info *get_frame_info(struct sk_buff *skb)
 {
 	struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
@@ -170,12 +187,51 @@ static struct ath_atx_tid *
 ath_get_skb_tid(struct ath_softc *sc, struct ath_node *an, struct sk_buff *skb)
 {
 	u8 tidno = skb->priority & IEEE80211_QOS_CTL_TID_MASK;
-	return ATH_AN_2_TID(an, tidno);
+	return ath_an_2_tid(an, tidno);
 }
 
+static void ath_swq_pull(struct ath_atx_tid *tid)
+{
+	struct sk_buff *skb;
+	struct ath_tx_control txctl;
+	struct ath_frame_info *fi;
+	int r;
+
+	if (!skb_queue_empty(&tid->i_q))
+		return;
+
+	if (!tid->swq_nonempty)
+		return;
+
+	skb = ieee80211_tx_dequeue(tid->an->sc->hw, tid->swq);
+	if (!skb) {
+		tid->swq_nonempty = false;
+	} else {
+		/* sad to do all this with axq_lock held */
+		memset(&txctl, 0, sizeof txctl);
+		txctl.txq = tid->txq;
+		txctl.sta = tid->an->sta;
+		r = ath_tx_prepare(tid->an->sc->hw, skb, &txctl);
+		if (WARN_ON(r != 0)) {
+			/** should not happen ??? */
+		} else {
+			/* perhaps not needed here ??? */
+			fi = get_frame_info(skb);
+			fi->txq = skb_get_queue_mapping(skb);
+
+			__skb_queue_tail(&tid->i_q, skb);
+			++tid->txq->pending_frames;
+		}
+	}
+ }
+
+
 static bool ath_tid_has_buffered(struct ath_atx_tid *tid)
 {
-	return !skb_queue_empty(&tid->buf_q) || !skb_queue_empty(&tid->retry_q);
+	if (!skb_queue_empty(&tid->buf_q) || !skb_queue_empty(&tid->retry_q) || !skb_queue_empty(&tid->i_q))
+		return true;
+	ath_swq_pull(tid);
+	return !skb_queue_empty(&tid->i_q);
 }
 
 static struct sk_buff *ath_tid_dequeue(struct ath_atx_tid *tid)
@@ -185,6 +241,12 @@ static struct sk_buff *ath_tid_dequeue(struct ath_atx_tid *tid)
 	skb = __skb_dequeue(&tid->retry_q);
 	if (!skb)
 		skb = __skb_dequeue(&tid->buf_q);
+	if (!skb)
+		skb = __skb_dequeue(&tid->i_q);
+	if (!skb) {
+		ath_swq_pull(tid);
+		skb = __skb_dequeue(&tid->i_q);
+	}
 
 	return skb;
 }
@@ -870,6 +932,10 @@ ath_tx_get_tid_subframe(struct ath_softc *sc, struct ath_txq *txq,
 		*q = &tid->retry_q;
 		if (skb_queue_empty(*q))
 			*q = &tid->buf_q;
+		if (skb_queue_empty(*q))
+			*q = &tid->i_q;
+		if (skb_queue_empty(*q))
+			ath_swq_pull(tid);
 
 		skb = skb_peek(*q);
 		if (!skb)
@@ -1482,7 +1548,7 @@ int ath_tx_aggr_start(struct ath_softc *sc, struct ieee80211_sta *sta,
 	ath_dbg(common, XMIT, "%s called\n", __func__);
 
 	an = (struct ath_node *)sta->drv_priv;
-	txtid = ATH_AN_2_TID(an, tid);
+	txtid = ath_an_2_tid(an, tid);
 	txq = txtid->txq;
 
 	ath_txq_lock(sc, txq);
@@ -1517,7 +1583,7 @@ void ath_tx_aggr_stop(struct ath_softc *sc, struct ieee80211_sta *sta, u16 tid)
 {
 	struct ath_common *common = ath9k_hw_common(sc->sc_ah);
 	struct ath_node *an = (struct ath_node *)sta->drv_priv;
-	struct ath_atx_tid *txtid = ATH_AN_2_TID(an, tid);
+	struct ath_atx_tid *txtid = ath_an_2_tid(an, tid);
 	struct ath_txq *txq = txtid->txq;
 
 	ath_dbg(common, XMIT, "%s called\n", __func__);
@@ -1533,6 +1599,7 @@ void ath_tx_aggr_sleep(struct ieee80211_sta *sta, struct ath_softc *sc,
 		       struct ath_node *an)
 {
 	struct ath_common *common = ath9k_hw_common(sc->sc_ah);
+	struct ieee80211_txq *swq;
 	struct ath_atx_tid *tid;
 	struct ath_txq *txq;
 	bool buffered;
@@ -1540,9 +1607,11 @@ void ath_tx_aggr_sleep(struct ieee80211_sta *sta, struct ath_softc *sc,
 
 	ath_dbg(common, XMIT, "%s called\n", __func__);
 
-	for (tidno = 0, tid = &an->tid[tidno];
-	     tidno < IEEE80211_NUM_TIDS; tidno++, tid++) {
+	for (tidno = 0;
+	     tidno < IEEE80211_NUM_TIDS; tidno++) {
 
+		swq = an->sta->txq[tidno];
+		tid = (struct ath_atx_tid *) swq->drv_priv;
 		txq = tid->txq;
 
 		ath_txq_lock(sc, txq);
@@ -1565,15 +1634,18 @@ void ath_tx_aggr_sleep(struct ieee80211_sta *sta, struct ath_softc *sc,
 void ath_tx_aggr_wakeup(struct ath_softc *sc, struct ath_node *an)
 {
 	struct ath_common *common = ath9k_hw_common(sc->sc_ah);
+	struct ieee80211_txq *swq;
 	struct ath_atx_tid *tid;
 	struct ath_txq *txq;
 	int tidno;
 
 	ath_dbg(common, XMIT, "%s called\n", __func__);
 
-	for (tidno = 0, tid = &an->tid[tidno];
-	     tidno < IEEE80211_NUM_TIDS; tidno++, tid++) {
+	for (tidno = 0;
+	     tidno < IEEE80211_NUM_TIDS; tidno++) {
 
+		swq = an->sta->txq[tidno];
+		tid = (struct ath_atx_tid *) swq->drv_priv;
 		txq = tid->txq;
 
 		ath_txq_lock(sc, txq);
@@ -1599,7 +1671,7 @@ void ath_tx_aggr_resume(struct ath_softc *sc, struct ieee80211_sta *sta,
 	ath_dbg(common, XMIT, "%s called\n", __func__);
 
 	an = (struct ath_node *)sta->drv_priv;
-	tid = ATH_AN_2_TID(an, tidno);
+	tid = ath_an_2_tid(an, tidno);
 	txq = tid->txq;
 
 	ath_txq_lock(sc, txq);
@@ -1637,7 +1709,7 @@ void ath9k_release_buffered_frames(struct ieee80211_hw *hw,
 		if (!(tids & 1))
 			continue;
 
-		tid = ATH_AN_2_TID(an, i);
+		tid = ath_an_2_tid(an, i);
 
 		ath_txq_lock(sc, tid->txq);
 		while (nframes > 0) {
@@ -2853,12 +2925,18 @@ int ath_tx_init(struct ath_softc *sc, int nbufs)
 
 void ath_tx_node_init(struct ath_softc *sc, struct ath_node *an)
 {
+	struct ieee80211_txq *swq;
+	struct ieee80211_sta *sta = an->sta;
+	struct ieee80211_vif *vif = an->vif;
 	struct ath_atx_tid *tid;
 	int tidno, acno;
 
-	for (tidno = 0, tid = &an->tid[tidno];
+	for (tidno = 0;
 	     tidno < IEEE80211_NUM_TIDS;
-	     tidno++, tid++) {
+	     tidno++) {
+		swq = sta ? sta->txq[tidno] : vif->txq;
+		tid = (struct ath_atx_tid *) swq->drv_priv;
+		tid->swq       = swq;
 		tid->an        = an;
 		tid->tidno     = tidno;
 		tid->seq_start = tid->seq_next = 0;
@@ -2866,23 +2944,33 @@ void ath_tx_node_init(struct ath_softc *sc, struct ath_node *an)
 		tid->baw_head  = tid->baw_tail = 0;
 		tid->active	   = false;
 		tid->clear_ps_filter = true;
+		tid->swq_nonempty  = false;
+		__skb_queue_head_init(&tid->i_q);
 		__skb_queue_head_init(&tid->buf_q);
 		__skb_queue_head_init(&tid->retry_q);
 		INIT_LIST_HEAD(&tid->list);
 		acno = TID_TO_WME_AC(tidno);
 		tid->txq = sc->tx.txq_map[acno];
+
+		if (!sta)
+			break; /* just one multicast ath_atx_tid */
 	}
 }
 
 void ath_tx_node_cleanup(struct ath_softc *sc, struct ath_node *an)
 {
+	struct ieee80211_txq *swq;
+	struct ieee80211_sta *sta = an->sta;
+	struct ieee80211_vif *vif = an->vif;
 	struct ath_atx_tid *tid;
 	struct ath_txq *txq;
 	int tidno;
 
-	for (tidno = 0, tid = &an->tid[tidno];
-	     tidno < IEEE80211_NUM_TIDS; tidno++, tid++) {
+	for (tidno = 0;
+	     tidno < IEEE80211_NUM_TIDS; tidno++) {
 
+		swq = sta ? sta->txq[tidno] : vif->txq;
+		tid = (struct ath_atx_tid *) swq->drv_priv;
 		txq = tid->txq;
 
 		ath_txq_lock(sc, txq);
@@ -2894,6 +2982,9 @@ void ath_tx_node_cleanup(struct ath_softc *sc, struct ath_node *an)
 		tid->active = false;
 
 		ath_txq_unlock(sc, txq);
+
+		if (!sta)
+			break; /* just one multicast ath_atx_tid */
 	}
 }
 
-- 
2.8.3