ath9k: Fix TX hang issue with Atheros chipsets

The hardware doesn't generate interrupts in some cases and so work
around this by monitoring the TX status periodically and reset the
chip if required.

This behavior of the hardware not generating the TX interrupts can
be noticed through ath9k debugfs interrupt statistics when heavy
traffic is being sent from STA to AP. One can easily see this behavior
when the STA is transmitting at a higher rates. The interrupt statistics
in the debugfs interface clearly shows that only RX interrupts alone
being generated and TX being stuck.

TX should be monitored through a timer and reset the chip only when
frames are queued to the hardware but TX interrupts are not generated
for the same even after one second. Also, we shouldn't remove holding
descriptor from AC queue if it happens to be the only descriptor and
schedule TX aggregation regarless of queue depth as it improves
scheduling of AMPDUs from software to hardware queue.

Signed-off-by: Vasanthakumar Thiagarajan <vasanth@atheros.com>
Signed-off-by: Senthil Balasubramanian <senthilkumar@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

1 files changed, 43 insertions, 14 deletions

diff --git a/drivers/net/wireless/ath/ath9k/xmit.c b/drivers/net/wireless/ath/ath9k/xmit.c
index 5de9878d2c12..a3bc4310a67c 100644
--- a/drivers/net/wireless/ath/ath9k/xmit.c
+++ b/drivers/net/wireless/ath/ath9k/xmit.c
@@ -857,6 +857,7 @@ struct ath_txq *ath_txq_setup(struct ath_softc *sc, int qtype, int subtype)
                 txq->axq_aggr_depth = 0;
                 txq->axq_totalqueued = 0;
                 txq->axq_linkbuf = NULL;
+                txq->axq_tx_inprogress = false;
                 sc->tx.txqsetup |= 1<<qnum;
         }
         return &sc->tx.txq[qnum];
@@ -1023,6 +1024,10 @@ void ath_draintxq(struct ath_softc *sc, struct ath_txq *txq, bool retry_tx)
                         ath_tx_complete_buf(sc, bf, &bf_head, 0, 0);
         }
 
+        spin_lock_bh(&txq->axq_lock);
+        txq->axq_tx_inprogress = false;
+        spin_unlock_bh(&txq->axq_lock);
+
         /* flush any pending frames if aggregation is enabled */
         if (sc->sc_flags & SC_OP_TXAGGR) {
                 if (!retry_tx) {
@@ -1103,8 +1108,7 @@ void ath_txq_schedule(struct ath_softc *sc, struct ath_txq *txq)
                 if (tid->paused)
                         continue;
 
-                if ((txq->axq_depth % 2) == 0)
-                        ath_tx_sched_aggr(sc, txq, tid);
+                ath_tx_sched_aggr(sc, txq, tid);
 
                 /*
                  * add tid to round-robin queue if more frames
@@ -1947,19 +1951,7 @@ static void ath_tx_processq(struct ath_softc *sc, struct ath_txq *txq)
                 if (bf->bf_stale) {
                         bf_held = bf;
                         if (list_is_last(&bf_held->list, &txq->axq_q)) {
-                                txq->axq_link = NULL;
-                                txq->axq_linkbuf = NULL;
                                 spin_unlock_bh(&txq->axq_lock);
-
-                                /*
-                                 * The holding descriptor is the last
-                                 * descriptor in queue. It's safe to remove
-                                 * the last holding descriptor in BH context.
-                                 */
-                                spin_lock_bh(&sc->tx.txbuflock);
-                                list_move_tail(&bf_held->list, &sc->tx.txbuf);
-                                spin_unlock_bh(&sc->tx.txbuflock);
-
                                 break;
                         } else {
                                 bf = list_entry(bf_held->list.next,
@@ -1996,6 +1988,7 @@ static void ath_tx_processq(struct ath_softc *sc, struct ath_txq *txq)
                         txq->axq_aggr_depth--;
 
                 txok = (ds->ds_txstat.ts_status == 0);
+                txq->axq_tx_inprogress = false;
                 spin_unlock_bh(&txq->axq_lock);
 
                 if (bf_held) {
@@ -2029,6 +2022,40 @@ static void ath_tx_processq(struct ath_softc *sc, struct ath_txq *txq)
         }
 }
 
+void ath_tx_complete_poll_work(struct work_struct *work)
+{
+        struct ath_softc *sc = container_of(work, struct ath_softc,
+                        tx_complete_work.work);
+        struct ath_txq *txq;
+        int i;
+        bool needreset = false;
+
+        for (i = 0; i < ATH9K_NUM_TX_QUEUES; i++)
+                if (ATH_TXQ_SETUP(sc, i)) {
+                        txq = &sc->tx.txq[i];
+                        spin_lock_bh(&txq->axq_lock);
+                        if (txq->axq_depth) {
+                                if (txq->axq_tx_inprogress) {
+                                        needreset = true;
+                                        spin_unlock_bh(&txq->axq_lock);
+                                        break;
+                                } else {
+                                        txq->axq_tx_inprogress = true;
+                                }
+                        }
+                        spin_unlock_bh(&txq->axq_lock);
+                }
+
+        if (needreset) {
+                DPRINTF(sc, ATH_DBG_RESET, "tx hung, resetting the chip\n");
+                ath_reset(sc, false);
+        }
+
+        queue_delayed_work(sc->hw->workqueue, &sc->tx_complete_work,
+                        msecs_to_jiffies(ATH_TX_COMPLETE_POLL_INT));
+}
+
+
 
 void ath_tx_tasklet(struct ath_softc *sc)
 {
@@ -2069,6 +2096,8 @@ int ath_tx_init(struct ath_softc *sc, int nbufs)
                 goto err;
         }
 
+        INIT_DELAYED_WORK(&sc->tx_complete_work, ath_tx_complete_poll_work);
+
 err:
         if (error != 0)
                 ath_tx_cleanup(sc);