NativeWifi tx mode expects QoS Data frames to be
delivered as Data frames with QoS part (e.g. tid)
being delievered out-of-band in fw tx command.
The QoS bit wasn't stripped before submitting to
firmware.
Stripping fixes two known problems:
* qca6174 IOT with some APs, e.g.
Cisco AIR-AP 1252 (which would crash after
ath10k association). Some ath9k APs would
crash as well.
* sniffing own tx frames via radiotap because,
e.g. wireshark was seeing QoS bit set but
since QoS Control was stripped in ath10k it
would parse beginning of LLC/SNAP
>From debugability point of view this removes the
ability to distinguish QoS from non-QoS frames
when sniffing own tx via radiotap. On the other
hand frames can be now parsed correctly without
special software modification.
Signed-off-by: Michal Kazior <[email protected]>
---
drivers/net/wireless/ath/ath10k/mac.c | 13 ++++++-------
1 file changed, 6 insertions(+), 7 deletions(-)
diff --git a/drivers/net/wireless/ath/ath10k/mac.c b/drivers/net/wireless/ath/ath10k/mac.c
index 1138f4f..625e2cb 100644
--- a/drivers/net/wireless/ath/ath10k/mac.c
+++ b/drivers/net/wireless/ath/ath10k/mac.c
@@ -2463,16 +2463,15 @@ static void ath10k_tx_h_nwifi(struct ieee80211_hw *hw, struct sk_buff *skb)
skb->data, (void *)qos_ctl - (void *)skb->data);
skb_pull(skb, IEEE80211_QOS_CTL_LEN);
- /* Fw/Hw generates a corrupted QoS Control Field for QoS NullFunc
- * frames. Powersave is handled by the fw/hw so QoS NyllFunc frames are
- * used only for CQM purposes (e.g. hostapd station keepalive ping) so
- * it is safe to downgrade to NullFunc.
+ /* Some firmware revisions don't handle sending QoS NullFunc well.
+ * These frames are mainly used for CQM purposes so it doesn't really
+ * matter whether QoS NullFunc or NullFunc are sent.
*/
hdr = (void *)skb->data;
- if (ieee80211_is_qos_nullfunc(hdr->frame_control)) {
- hdr->frame_control &= ~__cpu_to_le16(IEEE80211_STYPE_QOS_DATA);
+ if (ieee80211_is_qos_nullfunc(hdr->frame_control))
cb->htt.tid = HTT_DATA_TX_EXT_TID_NON_QOS_MCAST_BCAST;
- }
+
+ hdr->frame_control &= ~__cpu_to_le16(IEEE80211_STYPE_QOS_DATA);
}
static void ath10k_tx_h_add_p2p_noa_ie(struct ath10k *ar,
--
1.8.5.3
Michal Kazior <[email protected]> writes:
> NativeWifi tx mode expects QoS Data frames to be
> delivered as Data frames with QoS part (e.g. tid)
> being delievered out-of-band in fw tx command.
>
> The QoS bit wasn't stripped before submitting to
> firmware.
>
> Stripping fixes two known problems:
>
> * qca6174 IOT with some APs, e.g.
> Cisco AIR-AP 1252 (which would crash after
> ath10k association). Some ath9k APs would
> crash as well.
>
> * sniffing own tx frames via radiotap because,
> e.g. wireshark was seeing QoS bit set but
> since QoS Control was stripped in ath10k it
> would parse beginning of LLC/SNAP
>
>>From debugability point of view this removes the
> ability to distinguish QoS from non-QoS frames
> when sniffing own tx via radiotap. On the other
> hand frames can be now parsed correctly without
> special software modification.
>
> Signed-off-by: Michal Kazior <[email protected]>
Thanks, applied.
--
Kalle Valo
On Wed, 2015-03-11 at 14:25 +0100, Michal Kazior wrote:
> NativeWifi tx mode expects QoS Data frames to be
> delivered as Data frames with QoS part (e.g. tid)
> being delievered out-of-band in fw tx command.
>
> The QoS bit wasn't stripped before submitting to
> firmware.
>
> Stripping fixes two known problems:
>
> * qca6174 IOT with some APs, e.g.
> Cisco AIR-AP 1252 (which would crash after
> ath10k association). Some ath9k APs would
> crash as well.
It would probably be interesting to figure out why and fix that - this
is clearly a major issue.
johannes
On 11 March 2015 at 17:22, Johannes Berg <[email protected]> wrote:
> On Wed, 2015-03-11 at 14:25 +0100, Michal Kazior wrote:
>> NativeWifi tx mode expects QoS Data frames to be
>> delivered as Data frames with QoS part (e.g. tid)
>> being delievered out-of-band in fw tx command.
>>
>> The QoS bit wasn't stripped before submitting to
>> firmware.
>>
>> Stripping fixes two known problems:
>>
>> * qca6174 IOT with some APs, e.g.
>> Cisco AIR-AP 1252 (which would crash after
>> ath10k association). Some ath9k APs would
>> crash as well.
>
> It would probably be interesting to figure out why and fix that - this
> is clearly a major issue.
Good point. The patch was originally just a small fix for sniffing but
it happened to fix the IOT problem that we started seeing with the new
chip fw as well.
I believe that if 11n was disabled on APs the problem did not
reproduce. I think ath9k was crashing somewhere along the BA session
teardown (as per kernel call trace). The ath9k codebase from what I
know was rather old (some custom 3.4 fork) so perhaps it was a race of
some sort that has been fixed long since.
I'll try to get some sniff logs and take a look at this more closely.
Michał