summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authoradrian <adrian@FreeBSD.org>2013-05-08 21:23:51 +0000
committeradrian <adrian@FreeBSD.org>2013-05-08 21:23:51 +0000
commit04638b894ad1babedd83be1fe11700d4c5166eae (patch)
treeb962572ae00f3ec749787560f9f7d47e545dfb6e
parent0e97770857903861d2c055094e0d6f4872970f13 (diff)
downloadFreeBSD-src-04638b894ad1babedd83be1fe11700d4c5166eae.zip
FreeBSD-src-04638b894ad1babedd83be1fe11700d4c5166eae.tar.gz
Fix the holding descriptor logic to actually be "right" (for values
of "right".) Flip back on the "always continue TX DMA using the holding descriptor" code - by always setting ATH_BUF_BUSY and never setting axq_link to NULL. Since the holding descriptor is accessed via txq->axq_link and _that_ is done behind the TXQ lock rather than the TX path lock, the holding descriptor stuff itself needs to be behind the TXQ lock. So, do the mental gymnastics needed to do this. I've not seen any of the hardware failures that I was seeing when I last tried to do this. Tested: * AR5416, STA mode
-rw-r--r--sys/dev/ath/if_ath.c59
-rw-r--r--sys/dev/ath/if_athvar.h4
2 files changed, 55 insertions, 8 deletions
diff --git a/sys/dev/ath/if_ath.c b/sys/dev/ath/if_ath.c
index a71b3f8..6f5fb45 100644
--- a/sys/dev/ath/if_ath.c
+++ b/sys/dev/ath/if_ath.c
@@ -3896,6 +3896,7 @@ ath_tx_process_buf_completion(struct ath_softc *sc, struct ath_txq *txq,
struct ath_node *an = NULL;
ATH_TX_UNLOCK_ASSERT(sc);
+ ATH_TXQ_UNLOCK_ASSERT(txq);
/* If unicast frame, update general statistics */
if (ni != NULL) {
@@ -4000,6 +4001,28 @@ ath_tx_processq(struct ath_softc *sc, struct ath_txq *txq, int dosched)
break;
}
ATH_TXQ_REMOVE(txq, bf, bf_list);
+
+ /*
+ * Sanity check.
+ */
+ if (txq->axq_qnum != bf->bf_state.bfs_tx_queue) {
+ device_printf(sc->sc_dev,
+ "%s: TXQ=%d: bf=%p, bfs_tx_queue=%d\n",
+ __func__,
+ txq->axq_qnum,
+ bf,
+ bf->bf_state.bfs_tx_queue);
+ }
+ if (txq->axq_qnum != bf->bf_last->bf_state.bfs_tx_queue) {
+ device_printf(sc->sc_dev,
+ "%s: TXQ=%d: bf_last=%p, bfs_tx_queue=%d\n",
+ __func__,
+ txq->axq_qnum,
+ bf->bf_last,
+ bf->bf_last->bf_state.bfs_tx_queue);
+ }
+
+#if 0
if (txq->axq_depth > 0) {
/*
* More frames follow. Mark the buffer busy
@@ -4013,6 +4036,9 @@ ath_tx_processq(struct ath_softc *sc, struct ath_txq *txq, int dosched)
bf->bf_last->bf_flags |= ATH_BUF_BUSY;
} else
txq->axq_link = NULL;
+#else
+ bf->bf_last->bf_flags |= ATH_BUF_BUSY;
+#endif
if (bf->bf_state.bfs_aggr)
txq->axq_aggr_depth--;
@@ -4288,13 +4314,18 @@ ath_returnbuf_head(struct ath_softc *sc, struct ath_buf *bf)
void
ath_txq_freeholdingbuf(struct ath_softc *sc, struct ath_txq *txq)
{
- ATH_TXBUF_LOCK_ASSERT(sc);
+ ATH_TXBUF_UNLOCK_ASSERT(sc);
+ ATH_TXQ_LOCK_ASSERT(txq);
if (txq->axq_holdingbf == NULL)
return;
txq->axq_holdingbf->bf_flags &= ~ATH_BUF_BUSY;
+
+ ATH_TXBUF_LOCK(sc);
ath_returnbuf_tail(sc, txq->axq_holdingbf);
+ ATH_TXBUF_UNLOCK(sc);
+
txq->axq_holdingbf = NULL;
}
@@ -4307,7 +4338,10 @@ ath_txq_addholdingbuf(struct ath_softc *sc, struct ath_buf *bf)
{
struct ath_txq *txq;
- ATH_TXBUF_LOCK_ASSERT(sc);
+ txq = &sc->sc_txq[bf->bf_state.bfs_tx_queue];
+
+ ATH_TXBUF_UNLOCK_ASSERT(sc);
+ ATH_TXQ_LOCK_ASSERT(txq);
/* XXX assert ATH_BUF_BUSY is set */
@@ -4321,7 +4355,6 @@ ath_txq_addholdingbuf(struct ath_softc *sc, struct ath_buf *bf)
ath_returnbuf_tail(sc, bf);
return;
}
- txq = &sc->sc_txq[bf->bf_state.bfs_tx_queue];
ath_txq_freeholdingbuf(sc, txq);
txq->axq_holdingbf = bf;
}
@@ -4336,20 +4369,30 @@ ath_txq_addholdingbuf(struct ath_softc *sc, struct ath_buf *bf)
* for restart (eg for TDMA.)
*
* The caller must free the mbuf and recycle the node reference.
+ *
+ * XXX This method of handling busy / holding buffers is insanely stupid.
+ * It requires bf_state.bfs_tx_queue to be correctly assigned. It would
+ * be much nicer if buffers in the processq() methods would instead be
+ * always completed there (pushed onto a txq or ath_bufhead) so we knew
+ * exactly what hardware queue they came from in the first place.
*/
void
ath_freebuf(struct ath_softc *sc, struct ath_buf *bf)
{
+ struct ath_txq *txq;
+
+ txq = &sc->sc_txq[bf->bf_state.bfs_tx_queue];
+
KASSERT((bf->bf_node == NULL), ("%s: bf->bf_node != NULL\n", __func__));
KASSERT((bf->bf_m == NULL), ("%s: bf->bf_m != NULL\n", __func__));
/*
- * If this buffer is busy, push it onto the holding queue
+ * If this buffer is busy, push it onto the holding queue.
*/
if (bf->bf_flags & ATH_BUF_BUSY) {
- ATH_TXBUF_LOCK(sc);
+ ATH_TXQ_LOCK(txq);
ath_txq_addholdingbuf(sc, bf);
- ATH_TXBUF_UNLOCK(sc);
+ ATH_TXQ_UNLOCK(txq);
return;
}
@@ -4521,9 +4564,9 @@ ath_tx_draintxq(struct ath_softc *sc, struct ath_txq *txq)
/*
* Free the holding buffer if it exists
*/
- ATH_TXBUF_LOCK(sc);
+ ATH_TXQ_LOCK(txq);
ath_txq_freeholdingbuf(sc, txq);
- ATH_TXBUF_UNLOCK(sc);
+ ATH_TXQ_UNLOCK(txq);
/*
* Drain software queued frames which are on
diff --git a/sys/dev/ath/if_athvar.h b/sys/dev/ath/if_athvar.h
index f2b9230..662382f 100644
--- a/sys/dev/ath/if_athvar.h
+++ b/sys/dev/ath/if_athvar.h
@@ -381,6 +381,8 @@ struct ath_txq {
#define ATH_TXQ_LOCK(_tq) mtx_lock(&(_tq)->axq_lock)
#define ATH_TXQ_UNLOCK(_tq) mtx_unlock(&(_tq)->axq_lock)
#define ATH_TXQ_LOCK_ASSERT(_tq) mtx_assert(&(_tq)->axq_lock, MA_OWNED)
+#define ATH_TXQ_UNLOCK_ASSERT(_tq) mtx_assert(&(_tq)->axq_lock, \
+ MA_NOTOWNED)
#define ATH_NODE_LOCK(_an) mtx_lock(&(_an)->an_mtx)
@@ -964,6 +966,8 @@ struct ath_softc {
#define ATH_TXBUF_UNLOCK(_sc) mtx_unlock(&(_sc)->sc_txbuflock)
#define ATH_TXBUF_LOCK_ASSERT(_sc) \
mtx_assert(&(_sc)->sc_txbuflock, MA_OWNED)
+#define ATH_TXBUF_UNLOCK_ASSERT(_sc) \
+ mtx_assert(&(_sc)->sc_txbuflock, MA_NOTOWNED)
#define ATH_TXSTATUS_LOCK_INIT(_sc) do { \
snprintf((_sc)->sc_txcompname, sizeof((_sc)->sc_txcompname), \
OpenPOWER on IntegriCloud