Re: ath9k and pktgen generate WARNings.

[Helmut Schaa <helmut.schaa@googlemail.com>]

Am Donnerstag, 10. Februar 2011 schrieb Felix Fietkau:
> On 2011-02-10 7:00 PM, Ben Greear wrote:
> > On 02/10/2011 09:54 AM, Helmut Schaa wrote:
> >> Am Donnerstag, 10. Februar 2011 schrieb Ben Greear:
> >>> On 02/10/2011 12:31 AM, Helmut Schaa wrote:
> >>>> Am Donnerstag, 10. Februar 2011 schrieb Ben Greear:
> >>>>> I see this warning when I generate traffic with a hacked version
> >>>>> of pktgen.  This works with various other interfaces w/out problems,
> >>>>> so I think it's probably an ath9k and/or mac80211 bug.
> >>>>>
> >>>>>
> >>>>> WARNING: at /home/greearb/git/linux.wireless-testing-ct/drivers/net/wireless/ath/ath9k/xmit.c:1735 ath_tx_start+0x43c/0x607 [ath9k]()
> >>>>> Hardware name: To Be Filled By O.E.M.
> >>>>> Modules linked in: bridge nfs lockd bluetooth cryptd aes_i586 aes_generic veth 8021q garp stp llc fuse macvlan pktgen coretemp hwmon fscache nfs_acl auth_rpcgss
> >>>>> sunrpc ipv6 uinput arc4 ecb snd_hda_codec_realtek ath9k snd_hda_intel snd_hda_codec mac80211 snd_hwdep snd_seq snd_seq_device ath9k_common ath9k_hw snd_pcm ath
> >>>>> cfg80211 microcode snd_timer iTCO_wdt snd iTCO_vendor_support i2c_i801 serio_raw pcspkr soundcore r8169 snd_page_alloc mii i915 drm_kms_helper drm i2c_algo_bit
> >>>>> video [last unloaded: lockd]
> >>>>> Pid: 1729, comm: kpktgend_0 Tainted: G        W   2.6.38-rc4-wl+ #21
> >>>>> Call Trace:
> >>>>>     [<c043091b>] ? warn_slowpath_common+0x65/0x7a
> >>>>>     [<f8ded7e8>] ? ath_tx_start+0x43c/0x607 [ath9k]
> >>>>>     [<c043093f>] ? warn_slowpath_null+0xf/0x13
> >>>>>     [<f8ded7e8>] ? ath_tx_start+0x43c/0x607 [ath9k]
> >>>>>     [<f8de74d0>] ? ath9k_tx+0x14f/0x183 [ath9k]
> >>>>>     [<f8d1326d>] ? __ieee80211_tx+0x10c/0x18c [mac80211]
> >>>>>     [<f8d13397>] ? ieee80211_tx+0xaa/0x188 [mac80211]
> >>>>>     [<f8d135f3>] ? ieee80211_xmit+0x17e/0x186 [mac80211]
> >>>>>     [<f8d11cc0>] ? ieee80211_skb_resize+0x8e/0xd2 [mac80211]
> >>>>>     [<f8d1448b>] ? ieee80211_subif_start_xmit+0x643/0x65c [mac80211]
> >>>>>     [<c0440000>] ? rescuer_thread+0x25/0x1c8
> >>>>>     [<f92cd354>] ? pktgen_thread_worker+0x114c/0x1b44 [pktgen]
> >>>>>     [<f8d13e48>] ? ieee80211_subif_start_xmit+0x0/0x65c [mac80211]
> >>>>>     [<c042d612>] ? default_wake_function+0xb/0xd
> >>>>>     [<c04254c7>] ? __wake_up_common+0x34/0x5c
> >>>>>     [<c0443a29>] ? autoremove_wake_function+0x0/0x2f
> >>>>>     [<f92cc208>] ? pktgen_thread_worker+0x0/0x1b44 [pktgen]
> >>>>>     [<c044371a>] ? kthread+0x62/0x67
> >>>>>     [<c04436b8>] ? kthread+0x0/0x67
> >>>>>     [<c04035f6>] ? kernel_thread_helper+0x6/0x10
> >>>>>
> >>>>>
> >>>>> /* FIXME: tx power */
> >>>>> static void ath_tx_start_dma(struct ath_softc *sc, struct ath_buf *bf,
> >>>>> 			     struct ath_tx_control *txctl)
> >>>>> {
> >>>>> 	struct sk_buff *skb = bf->bf_mpdu;
> >>>>> 	struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
> >>>>> 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
> >>>>> 	struct list_head bf_head;
> >>>>> 	struct ath_atx_tid *tid = NULL;
> >>>>> 	u8 tidno;
> >>>>>
> >>>>> 	spin_lock_bh(&txctl->txq->axq_lock);
> >>>>>
> >>>>> 	if (ieee80211_is_data_qos(hdr->frame_control)&&   txctl->an) {
> >>>>> 		tidno = ieee80211_get_qos_ctl(hdr)[0]&
> >>>>> 			IEEE80211_QOS_CTL_TID_MASK;
> >>>>> 		tid = ATH_AN_2_TID(txctl->an, tidno);
> >>>>>
> >>>>> 		WARN_ON(tid->ac->txq != txctl->txq);
> >>>>> 	}
> >>>>>
> >>>>> Anyone have any ideas on this one?
> >>>>
> >>>> pktgen seems to directly inject frames via the xmit callback and hence
> >>>> mac80211's select_queue callback isn't used for proper queue assignment.
> >>>>
> >>>> However, you should be able to specify the queue manuelly with
> >>>> "cur_queue_map".
> >>>
> >>> Yes, but I don't think I should have to know this detail.
> >>>
> >>> Once you send pkts with queue-map of 0 to ath9k, it spews
> >>> kernel warnings and then you have to rmmod the NIC before
> >>> it will start working again (because it's queues are
> >>> messed up probably).
> >>>
> >>> I'm not sure my patch is the correct way to fix it, but
> >>> we need to do _something_ in mac80211 and/or ath9k I think.
> >>
> >> Why not make pktgen use the select_queue callback instead?
> > 
> > It's a test tool and should be able to force queue selection
> > if it wants to.  I'd rather not have special case code to
> > know if underlying interface is wifi (and thus need special
> > queue selection).
> > 
> > I'd rather carry that hack to mac80211 I posted instead,
> > if it comes to that.
> How about adding a patch that makes mac80211 detect a mismatch between
> skb->priority and the queue mapping and then adjust the skb priority
> based on that.

Sounds reasonable to me. But please also add a WARN_ON_ONCE in that case
as otherwise we wouldn't detect broken queue selection under "normal"
conditions anymore.

Helmut