Tong Zhu <zhutong@xxxxxxxxxx> writes:
> Hello
>
> After moving to kernel 4.14 from 4.1, our station with an AR93xx pcie card began
> to see packet loss during background scanning. After inspecting the changes, I
> realized ath9k no longer pushes to stop traffic at VIF for off-channel scan after
> this change.
>
> 50f08edf98096a68f01ff4566b605a25bf8e42ce ath9k: Switch to using mac80211 intermediate software queues.
>
> Here is a snippet of a ftrace function sequence during a scan.
>
> kworker/u4:0-28892 [001] .n..... 3082.957750: ieee80211_offchannel_stop_vifs <-ieee80211_scan_work
> XX-XXXXXXXXXXd-489 [001] .....12 3082.971105: __ieee80211_subif_start_xmit <-ieee80211_subif_start_xmit
> XX-XXXXXXXXXXd-489 [001] .....14 3082.971128: ath_tx_txqaddbuf <-ath_txq_schedule
> irq/175-ath9k-189 [001] .....15 3082.998888: ath_tx_process_buffer <-ath_tx_edma_tasklet
> irq/175-ath9k-189 [001] .....16 3082.998898: ath_tx_txqaddbuf <-ath_txq_schedule
> irq/175-ath9k-189 [000] .....15 3083.011497: ath_tx_process_buffer <-ath_tx_edma_tasklet
> irq/175-ath9k-189 [000] .....16 3083.011504: ath_tx_txqaddbuf <-ath_txq_schedule
> irq/175-ath9k-189 [001] .....15 3083.022261: ath_tx_process_buffer <-ath_tx_edma_tasklet
> irq/175-ath9k-189 [001] .....16 3083.022268: ath_tx_txqaddbuf <-ath_txq_schedule
> irq/175-ath9k-189 [000] .....15 3083.034131: ath_tx_process_buffer <-ath_tx_edma_tasklet
> irq/175-ath9k-189 [000] .....15 3083.034134: ath_tx_complete_buf <-ath_tx_process_buffer
> irq/175-ath9k-189 [000] .....15 3083.034136: ath_tx_complete <-ath_tx_complete_buf
> kworker/u4:0-28892 [001] ....... 3083.083246: ieee80211_offchannel_return <-ieee80211_scan_work
>
> Between timestamp 3082.957750 and 3083.083246, the device went off channel. A packet
> came down and reached hardware queue. Hardware TX processing continued. It is a
> serious issue if PDUs are indeed sent out during off-channel.
I believe this was fixed by this commit:
21a5d4c3a45c ("mac80211: add stop/start logic for software TXQs")
which first appeared in kernel 4.20. It doesn't appear to have been
backported to 4.14, so I suppose it makes sense if you're seeing queues
not getting stopped on that kernel...
-Toke
