ath9k: use split rx buffers to get rid of order-1 skb allocations

With this change, less CPU time is spent trying to look for consecutive
pages for rx skbs. This also reduces the socket memory required for IP/UDP
reassembly.
Only two buffers per frame are supported. Frames spanning more buffers
will be dropped, but the buffer size is enough to handle the required
AMSDU size.

Signed-off-by: Jouni Malinen <jouni.malinen@atheros.com>
Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

1 files changed, 64 insertions, 24 deletions

diff --git a/drivers/net/wireless/ath/ath9k/recv.c b/drivers/net/wireless/ath/ath9k/recv.c
index b2b12a293c70..daf171d2f610 100644
--- a/drivers/net/wireless/ath/ath9k/recv.c
+++ b/drivers/net/wireless/ath/ath9k/recv.c
@@ -209,11 +209,6 @@ static int ath_rx_edma_init(struct ath_softc *sc, int nbufs)
 	int error = 0, i;
 	u32 size;
 
-
-	common->rx_bufsize = roundup(IEEE80211_MAX_MPDU_LEN +
-				     ah->caps.rx_status_len,
-				     min(common->cachelsz, (u16)64));
-
 	ath9k_hw_set_rx_bufsize(ah, common->rx_bufsize -
 				    ah->caps.rx_status_len);
 
@@ -300,12 +295,12 @@ int ath_rx_init(struct ath_softc *sc, int nbufs)
 	sc->sc_flags &= ~SC_OP_RXFLUSH;
 	spin_lock_init(&sc->rx.rxbuflock);
 
+	common->rx_bufsize = IEEE80211_MAX_MPDU_LEN / 2 +
+			     sc->sc_ah->caps.rx_status_len;
+
 	if (sc->sc_ah->caps.hw_caps & ATH9K_HW_CAP_EDMA) {
 		return ath_rx_edma_init(sc, nbufs);
 	} else {
-		common->rx_bufsize = roundup(IEEE80211_MAX_MPDU_LEN,
-				min(common->cachelsz, (u16)64));
-
 		ath_dbg(common, ATH_DBG_CONFIG, "cachelsz %u rxbufsize %u\n",
 			common->cachelsz, common->rx_bufsize);
 
@@ -815,15 +810,9 @@ static bool ath9k_rx_accept(struct ath_common *common,
 	if (rx_stats->rs_datalen > (common->rx_bufsize - rx_status_len))
 		return false;
 
-	/*
-	 * rs_more indicates chained descriptors which can be used
-	 * to link buffers together for a sort of scatter-gather
-	 * operation.
-	 * reject the frame, we don't support scatter-gather yet and
-	 * the frame is probably corrupt anyway
-	 */
+	/* Only use error bits from the last fragment */
 	if (rx_stats->rs_more)
-		return false;
+		return true;
 
 	/*
 	 * The rx_stats->rs_status will not be set until the end of the
@@ -981,6 +970,10 @@ static int ath9k_rx_skb_preprocess(struct ath_common *common,
 	if (!ath9k_rx_accept(common, hdr, rx_status, rx_stats, decrypt_error))
 		return -EINVAL;
 
+	/* Only use status info from the last fragment */
+	if (rx_stats->rs_more)
+		return 0;
+
 	ath9k_process_rssi(common, hw, hdr, rx_stats);
 
 	if (ath9k_process_rate(common, hw, rx_stats, rx_status))
@@ -1582,7 +1575,7 @@ div_comb_done:
 int ath_rx_tasklet(struct ath_softc *sc, int flush, bool hp)
 {
 	struct ath_buf *bf;
-	struct sk_buff *skb = NULL, *requeue_skb;
+	struct sk_buff *skb = NULL, *requeue_skb, *hdr_skb;
 	struct ieee80211_rx_status *rxs;
 	struct ath_hw *ah = sc->sc_ah;
 	struct ath_common *common = ath9k_hw_common(ah);
@@ -1633,8 +1626,17 @@ int ath_rx_tasklet(struct ath_softc *sc, int flush, bool hp)
 		if (!skb)
 			continue;
 
-		hdr = (struct ieee80211_hdr *) (skb->data + rx_status_len);
-		rxs =  IEEE80211_SKB_RXCB(skb);
+		/*
+		 * Take frame header from the first fragment and RX status from
+		 * the last one.
+		 */
+		if (sc->rx.frag)
+			hdr_skb = sc->rx.frag;
+		else
+			hdr_skb = skb;
+
+		hdr = (struct ieee80211_hdr *) (hdr_skb->data + rx_status_len);
+		rxs = IEEE80211_SKB_RXCB(hdr_skb);
 
 		ath_debug_stat_rx(sc, &rs);
 
@@ -1643,12 +1645,12 @@ int ath_rx_tasklet(struct ath_softc *sc, int flush, bool hp)
 		 * chain it back at the queue without processing it.
 		 */
 		if (flush)
-			goto requeue;
+			goto requeue_drop_frag;
 
 		retval = ath9k_rx_skb_preprocess(common, hw, hdr, &rs,
 						 rxs, &decrypt_error);
 		if (retval)
-			goto requeue;
+			goto requeue_drop_frag;
 
 		rxs->mactime = (tsf & ~0xffffffffULL) | rs.rs_tstamp;
 		if (rs.rs_tstamp > tsf_lower &&
@@ -1668,7 +1670,7 @@ int ath_rx_tasklet(struct ath_softc *sc, int flush, bool hp)
 		 * skb and put it at the tail of the sc->rx.rxbuf list for
 		 * processing. */
 		if (!requeue_skb)
-			goto requeue;
+			goto requeue_drop_frag;
 
 		/* Unmap the frame */
 		dma_unmap_single(sc->dev, bf->bf_buf_addr,
@@ -1679,8 +1681,9 @@ int ath_rx_tasklet(struct ath_softc *sc, int flush, bool hp)
 		if (ah->caps.rx_status_len)
 			skb_pull(skb, ah->caps.rx_status_len);
 
-		ath9k_rx_skb_postprocess(common, skb, &rs,
-					 rxs, decrypt_error);
+		if (!rs.rs_more)
+			ath9k_rx_skb_postprocess(common, hdr_skb, &rs,
+						 rxs, decrypt_error);
 
 		/* We will now give hardware our shiny new allocated skb */
 		bf->bf_mpdu = requeue_skb;
@@ -1697,6 +1700,38 @@ int ath_rx_tasklet(struct ath_softc *sc, int flush, bool hp)
 			break;
 		}
 
+		if (rs.rs_more) {
+			/*
+			 * rs_more indicates chained descriptors which can be
+			 * used to link buffers together for a sort of
+			 * scatter-gather operation.
+			 */
+			if (sc->rx.frag) {
+				/* too many fragments - cannot handle frame */
+				dev_kfree_skb_any(sc->rx.frag);
+				dev_kfree_skb_any(skb);
+				skb = NULL;
+			}
+			sc->rx.frag = skb;
+			goto requeue;
+		}
+
+		if (sc->rx.frag) {
+			int space = skb->len - skb_tailroom(hdr_skb);
+
+			sc->rx.frag = NULL;
+
+			if (pskb_expand_head(hdr_skb, 0, space, GFP_ATOMIC) < 0) {
+				dev_kfree_skb(skb);
+				goto requeue_drop_frag;
+			}
+
+			skb_copy_from_linear_data(skb, skb_put(hdr_skb, skb->len),
+						  skb->len);
+			dev_kfree_skb_any(skb);
+			skb = hdr_skb;
+		}
+
 		/*
 		 * change the default rx antenna if rx diversity chooses the
 		 * other antenna 3 times in a row.
@@ -1722,6 +1757,11 @@ int ath_rx_tasklet(struct ath_softc *sc, int flush, bool hp)
 
 		ieee80211_rx(hw, skb);
 
+requeue_drop_frag:
+		if (sc->rx.frag) {
+			dev_kfree_skb_any(sc->rx.frag);
+			sc->rx.frag = NULL;
+		}
 requeue:
 		if (edma) {
 			list_add_tail(&bf->list, &sc->rx.rxbuf);