[PATCH 0/5] ath10k: tx performance patches

[PATCH 0/5] ath10k: tx performance patches

[Michal Kazior]

Hi,

This patchset reduces CPU load and improves
improves TX performance on AP135 by additional
30mbps (give or take; 560mbps -> 590mbps), at
least on my testbed.


Michal Kazior (5):
  ath10k: use num_pending_tx instead of msdu id bitmap
  ath10k: avoid needless memset on TX path
  ath10k: decouple HTT TX completions
  ath10k: cleanup HTT TX functions
  ath10k: use msdu headroom to store txfrag

 drivers/net/wireless/ath/ath10k/core.h   |   12 +-
 drivers/net/wireless/ath/ath10k/htc.c    |    2 +-
 drivers/net/wireless/ath/ath10k/htt_rx.c |    4 +-
 drivers/net/wireless/ath/ath10k/htt_tx.c |  249 ++++++++++++------------------
 drivers/net/wireless/ath/ath10k/mac.c    |    9 +-
 drivers/net/wireless/ath/ath10k/txrx.c   |   65 +++-----
 drivers/net/wireless/ath/ath10k/txrx.h   |    5 +-
 7 files changed, 131 insertions(+), 215 deletions(-)

-- 
1.7.9.5


_______________________________________________
ath10k mailing list
ath10k@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/ath10k

[PATCH 0/5] ath10k: tx performance patches

[Michal Kazior]

Hi,

This patchset reduces CPU load and improves
improves TX performance on AP135 by additional
30mbps (give or take; 560mbps -> 590mbps), at
least on my testbed.


Michal Kazior (5):
  ath10k: use num_pending_tx instead of msdu id bitmap
  ath10k: avoid needless memset on TX path
  ath10k: decouple HTT TX completions
  ath10k: cleanup HTT TX functions
  ath10k: use msdu headroom to store txfrag

 drivers/net/wireless/ath/ath10k/core.h   |   12 +-
 drivers/net/wireless/ath/ath10k/htc.c    |    2 +-
 drivers/net/wireless/ath/ath10k/htt_rx.c |    4 +-
 drivers/net/wireless/ath/ath10k/htt_tx.c |  249 ++++++++++++------------------
 drivers/net/wireless/ath/ath10k/mac.c    |    9 +-
 drivers/net/wireless/ath/ath10k/txrx.c   |   65 +++-----
 drivers/net/wireless/ath/ath10k/txrx.h   |    5 +-
 7 files changed, 131 insertions(+), 215 deletions(-)

-- 
1.7.9.5

[PATCH 1/5] ath10k: use num_pending_tx instead of msdu id bitmap

[Michal Kazior]

It's more efficient to simply check num_pending_tx
value instead of traversing whole bitmap of
msdu ids.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
---
 drivers/net/wireless/ath/ath10k/mac.c  |    3 +--
 drivers/net/wireless/ath/ath10k/txrx.c |    2 +-
 2 files changed, 2 insertions(+), 3 deletions(-)

diff --git a/drivers/net/wireless/ath/ath10k/mac.c b/drivers/net/wireless/ath/ath10k/mac.c
index 11aa13e..5a56833 100644
--- a/drivers/net/wireless/ath/ath10k/mac.c
+++ b/drivers/net/wireless/ath/ath10k/mac.c
@@ -2853,8 +2853,7 @@ static void ath10k_flush(struct ieee80211_hw *hw, u32 queues, bool drop)
 			bool empty;
 
 			spin_lock_bh(&ar->htt.tx_lock);
-			empty = bitmap_empty(ar->htt.used_msdu_ids,
-					     ar->htt.max_num_pending_tx);
+			empty = (ar->htt.num_pending_tx == 0);
 			spin_unlock_bh(&ar->htt.tx_lock);
 
 			skip = (ar->state == ATH10K_STATE_WEDGED);
diff --git a/drivers/net/wireless/ath/ath10k/txrx.c b/drivers/net/wireless/ath/ath10k/txrx.c
index 68b6fae..37b8196 100644
--- a/drivers/net/wireless/ath/ath10k/txrx.c
+++ b/drivers/net/wireless/ath/ath10k/txrx.c
@@ -96,7 +96,7 @@ exit:
 	htt->pending_tx[ATH10K_SKB_CB(txdesc)->htt.msdu_id] = NULL;
 	ath10k_htt_tx_free_msdu_id(htt, ATH10K_SKB_CB(txdesc)->htt.msdu_id);
 	__ath10k_htt_tx_dec_pending(htt);
-	if (bitmap_empty(htt->used_msdu_ids, htt->max_num_pending_tx))
+	if (htt->num_pending_tx == 0)
 		wake_up(&htt->empty_tx_wq);
 	spin_unlock_bh(&htt->tx_lock);
 
-- 
1.7.9.5


_______________________________________________
ath10k mailing list
ath10k@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/ath10k

[PATCH 1/5] ath10k: use num_pending_tx instead of msdu id bitmap

[Michal Kazior]

It's more efficient to simply check num_pending_tx
value instead of traversing whole bitmap of
msdu ids.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
---
 drivers/net/wireless/ath/ath10k/mac.c  |    3 +--
 drivers/net/wireless/ath/ath10k/txrx.c |    2 +-
 2 files changed, 2 insertions(+), 3 deletions(-)

diff --git a/drivers/net/wireless/ath/ath10k/mac.c b/drivers/net/wireless/ath/ath10k/mac.c
index 11aa13e..5a56833 100644
--- a/drivers/net/wireless/ath/ath10k/mac.c
+++ b/drivers/net/wireless/ath/ath10k/mac.c
@@ -2853,8 +2853,7 @@ static void ath10k_flush(struct ieee80211_hw *hw, u32 queues, bool drop)
 			bool empty;
 
 			spin_lock_bh(&ar->htt.tx_lock);
-			empty = bitmap_empty(ar->htt.used_msdu_ids,
-					     ar->htt.max_num_pending_tx);
+			empty = (ar->htt.num_pending_tx == 0);
 			spin_unlock_bh(&ar->htt.tx_lock);
 
 			skip = (ar->state == ATH10K_STATE_WEDGED);
diff --git a/drivers/net/wireless/ath/ath10k/txrx.c b/drivers/net/wireless/ath/ath10k/txrx.c
index 68b6fae..37b8196 100644
--- a/drivers/net/wireless/ath/ath10k/txrx.c
+++ b/drivers/net/wireless/ath/ath10k/txrx.c
@@ -96,7 +96,7 @@ exit:
 	htt->pending_tx[ATH10K_SKB_CB(txdesc)->htt.msdu_id] = NULL;
 	ath10k_htt_tx_free_msdu_id(htt, ATH10K_SKB_CB(txdesc)->htt.msdu_id);
 	__ath10k_htt_tx_dec_pending(htt);
-	if (bitmap_empty(htt->used_msdu_ids, htt->max_num_pending_tx))
+	if (htt->num_pending_tx == 0)
 		wake_up(&htt->empty_tx_wq);
 	spin_unlock_bh(&htt->tx_lock);
 
-- 
1.7.9.5

[PATCH 2/5] ath10k: avoid needless memset on TX path

[Michal Kazior]

This reduces number of memory accesses and
hopefully contributes to better performance in the
future.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
---
 drivers/net/wireless/ath/ath10k/htc.c    |    2 +-
 drivers/net/wireless/ath/ath10k/htt_tx.c |    4 +++-
 drivers/net/wireless/ath/ath10k/mac.c    |    4 +++-
 drivers/net/wireless/ath/ath10k/txrx.c   |    1 -
 4 files changed, 7 insertions(+), 4 deletions(-)

diff --git a/drivers/net/wireless/ath/ath10k/htc.c b/drivers/net/wireless/ath/ath10k/htc.c
index d0d7212..3118d75 100644
--- a/drivers/net/wireless/ath/ath10k/htc.c
+++ b/drivers/net/wireless/ath/ath10k/htc.c
@@ -103,10 +103,10 @@ static void ath10k_htc_prepare_tx_skb(struct ath10k_htc_ep *ep,
 	struct ath10k_htc_hdr *hdr;
 
 	hdr = (struct ath10k_htc_hdr *)skb->data;
-	memset(hdr, 0, sizeof(*hdr));
 
 	hdr->eid = ep->eid;
 	hdr->len = __cpu_to_le16(skb->len - sizeof(*hdr));
+	hdr->flags = 0;
 
 	spin_lock_bh(&ep->htc->tx_lock);
 	hdr->seq_no = ep->seq_no++;
diff --git a/drivers/net/wireless/ath/ath10k/htt_tx.c b/drivers/net/wireless/ath/ath10k/htt_tx.c
index 4548128..c4bbf74 100644
--- a/drivers/net/wireless/ath/ath10k/htt_tx.c
+++ b/drivers/net/wireless/ath/ath10k/htt_tx.c
@@ -384,9 +384,11 @@ int ath10k_htt_mgmt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 	/* refcount is decremented by HTC and HTT completions until it reaches
 	 * zero and is freed */
 	skb_cb = ATH10K_SKB_CB(txdesc);
+	skb_cb->htt.is_conf = false;
 	skb_cb->htt.msdu_id = msdu_id;
 	skb_cb->htt.refcount = 2;
 	skb_cb->htt.msdu = msdu;
+	skb_cb->htt.txfrag = NULL;
 
 	res = ath10k_htc_send(&htt->ar->htc, htt->eid, txdesc);
 	if (res)
@@ -505,7 +507,6 @@ int ath10k_htt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 
 	skb_put(txdesc, desc_len);
 	cmd = (struct htt_cmd *)txdesc->data;
-	memset(cmd, 0, desc_len);
 
 	tid = ATH10K_SKB_CB(msdu)->htt.tid;
 
@@ -555,6 +556,7 @@ int ath10k_htt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 	/* refcount is decremented by HTC and HTT completions until it reaches
 	 * zero and is freed */
 	skb_cb = ATH10K_SKB_CB(txdesc);
+	skb_cb->htt.is_conf = false;
 	skb_cb->htt.msdu_id = msdu_id;
 	skb_cb->htt.refcount = 2;
 	skb_cb->htt.txfrag = txfrag;
diff --git a/drivers/net/wireless/ath/ath10k/mac.c b/drivers/net/wireless/ath/ath10k/mac.c
index 5a56833..0d367e4 100644
--- a/drivers/net/wireless/ath/ath10k/mac.c
+++ b/drivers/net/wireless/ath/ath10k/mac.c
@@ -1757,7 +1757,9 @@ static void ath10k_tx(struct ieee80211_hw *hw,
 		ath10k_tx_h_seq_no(skb);
 	}
 
-	memset(ATH10K_SKB_CB(skb), 0, sizeof(*ATH10K_SKB_CB(skb)));
+	ATH10K_SKB_CB(skb)->is_mapped = false;
+	ATH10K_SKB_CB(skb)->is_aborted = false;
+	ATH10K_SKB_CB(skb)->htt.is_offchan = false;
 	ATH10K_SKB_CB(skb)->htt.vdev_id = vdev_id;
 	ATH10K_SKB_CB(skb)->htt.tid = tid;
 
diff --git a/drivers/net/wireless/ath/ath10k/txrx.c b/drivers/net/wireless/ath/ath10k/txrx.c
index 37b8196..f6fed31 100644
--- a/drivers/net/wireless/ath/ath10k/txrx.c
+++ b/drivers/net/wireless/ath/ath10k/txrx.c
@@ -75,7 +75,6 @@ void ath10k_txrx_tx_unref(struct ath10k_htt *htt, struct sk_buff *txdesc)
 	ath10k_report_offchan_tx(htt->ar, msdu);
 
 	info = IEEE80211_SKB_CB(msdu);
-	memset(&info->status, 0, sizeof(info->status));
 
 	if (ATH10K_SKB_CB(txdesc)->htt.discard) {
 		ieee80211_free_txskb(htt->ar->hw, msdu);
-- 
1.7.9.5


_______________________________________________
ath10k mailing list
ath10k@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/ath10k

[PATCH 2/5] ath10k: avoid needless memset on TX path

[Michal Kazior]

This reduces number of memory accesses and
hopefully contributes to better performance in the
future.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
---
 drivers/net/wireless/ath/ath10k/htc.c    |    2 +-
 drivers/net/wireless/ath/ath10k/htt_tx.c |    4 +++-
 drivers/net/wireless/ath/ath10k/mac.c    |    4 +++-
 drivers/net/wireless/ath/ath10k/txrx.c   |    1 -
 4 files changed, 7 insertions(+), 4 deletions(-)

diff --git a/drivers/net/wireless/ath/ath10k/htc.c b/drivers/net/wireless/ath/ath10k/htc.c
index d0d7212..3118d75 100644
--- a/drivers/net/wireless/ath/ath10k/htc.c
+++ b/drivers/net/wireless/ath/ath10k/htc.c
@@ -103,10 +103,10 @@ static void ath10k_htc_prepare_tx_skb(struct ath10k_htc_ep *ep,
 	struct ath10k_htc_hdr *hdr;
 
 	hdr = (struct ath10k_htc_hdr *)skb->data;
-	memset(hdr, 0, sizeof(*hdr));
 
 	hdr->eid = ep->eid;
 	hdr->len = __cpu_to_le16(skb->len - sizeof(*hdr));
+	hdr->flags = 0;
 
 	spin_lock_bh(&ep->htc->tx_lock);
 	hdr->seq_no = ep->seq_no++;
diff --git a/drivers/net/wireless/ath/ath10k/htt_tx.c b/drivers/net/wireless/ath/ath10k/htt_tx.c
index 4548128..c4bbf74 100644
--- a/drivers/net/wireless/ath/ath10k/htt_tx.c
+++ b/drivers/net/wireless/ath/ath10k/htt_tx.c
@@ -384,9 +384,11 @@ int ath10k_htt_mgmt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 	/* refcount is decremented by HTC and HTT completions until it reaches
 	 * zero and is freed */
 	skb_cb = ATH10K_SKB_CB(txdesc);
+	skb_cb->htt.is_conf = false;
 	skb_cb->htt.msdu_id = msdu_id;
 	skb_cb->htt.refcount = 2;
 	skb_cb->htt.msdu = msdu;
+	skb_cb->htt.txfrag = NULL;
 
 	res = ath10k_htc_send(&htt->ar->htc, htt->eid, txdesc);
 	if (res)
@@ -505,7 +507,6 @@ int ath10k_htt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 
 	skb_put(txdesc, desc_len);
 	cmd = (struct htt_cmd *)txdesc->data;
-	memset(cmd, 0, desc_len);
 
 	tid = ATH10K_SKB_CB(msdu)->htt.tid;
 
@@ -555,6 +556,7 @@ int ath10k_htt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 	/* refcount is decremented by HTC and HTT completions until it reaches
 	 * zero and is freed */
 	skb_cb = ATH10K_SKB_CB(txdesc);
+	skb_cb->htt.is_conf = false;
 	skb_cb->htt.msdu_id = msdu_id;
 	skb_cb->htt.refcount = 2;
 	skb_cb->htt.txfrag = txfrag;
diff --git a/drivers/net/wireless/ath/ath10k/mac.c b/drivers/net/wireless/ath/ath10k/mac.c
index 5a56833..0d367e4 100644
--- a/drivers/net/wireless/ath/ath10k/mac.c
+++ b/drivers/net/wireless/ath/ath10k/mac.c
@@ -1757,7 +1757,9 @@ static void ath10k_tx(struct ieee80211_hw *hw,
 		ath10k_tx_h_seq_no(skb);
 	}
 
-	memset(ATH10K_SKB_CB(skb), 0, sizeof(*ATH10K_SKB_CB(skb)));
+	ATH10K_SKB_CB(skb)->is_mapped = false;
+	ATH10K_SKB_CB(skb)->is_aborted = false;
+	ATH10K_SKB_CB(skb)->htt.is_offchan = false;
 	ATH10K_SKB_CB(skb)->htt.vdev_id = vdev_id;
 	ATH10K_SKB_CB(skb)->htt.tid = tid;
 
diff --git a/drivers/net/wireless/ath/ath10k/txrx.c b/drivers/net/wireless/ath/ath10k/txrx.c
index 37b8196..f6fed31 100644
--- a/drivers/net/wireless/ath/ath10k/txrx.c
+++ b/drivers/net/wireless/ath/ath10k/txrx.c
@@ -75,7 +75,6 @@ void ath10k_txrx_tx_unref(struct ath10k_htt *htt, struct sk_buff *txdesc)
 	ath10k_report_offchan_tx(htt->ar, msdu);
 
 	info = IEEE80211_SKB_CB(msdu);
-	memset(&info->status, 0, sizeof(info->status));
 
 	if (ATH10K_SKB_CB(txdesc)->htt.discard) {
 		ieee80211_free_txskb(htt->ar->hw, msdu);
-- 
1.7.9.5

[PATCH 3/5] ath10k: decouple HTT TX completions

[Michal Kazior]

Until now the all MSDU transfer related structures
were freed when all resources were unreferenced.

Now HTC transfer is freed independently and HTT
transfer is so too.

This yields a way more simpler ath10k_skb_cb and
should possibly enable parallel pipe processing
(which is now serialized in
ath10k_pci_process_ce routine).

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
---
 drivers/net/wireless/ath/ath10k/core.h   |    8 ----
 drivers/net/wireless/ath/ath10k/htt_rx.c |    4 +-
 drivers/net/wireless/ath/ath10k/htt_tx.c |   64 ++++--------------------------
 drivers/net/wireless/ath/ath10k/mac.c    |    2 -
 drivers/net/wireless/ath/ath10k/txrx.c   |   52 ++++++++----------------
 drivers/net/wireless/ath/ath10k/txrx.h   |    5 +--
 6 files changed, 27 insertions(+), 108 deletions(-)

diff --git a/drivers/net/wireless/ath/ath10k/core.h b/drivers/net/wireless/ath/ath10k/core.h
index fcf94ee..4563f80 100644
--- a/drivers/net/wireless/ath/ath10k/core.h
+++ b/drivers/net/wireless/ath/ath10k/core.h
@@ -52,18 +52,10 @@ struct ath10k_skb_cb {
 
 	struct {
 		u8 vdev_id;
-		u16 msdu_id;
 		u8 tid;
 		bool is_offchan;
-		bool is_conf;
-		bool discard;
-		bool no_ack;
-		u8 refcount;
 		struct sk_buff *txfrag;
-		struct sk_buff *msdu;
 	} __packed htt;
-
-	/* 4 bytes left on 64bit arch */
 } __packed;
 
 static inline struct ath10k_skb_cb *ATH10K_SKB_CB(struct sk_buff *skb)
diff --git a/drivers/net/wireless/ath/ath10k/htt_rx.c b/drivers/net/wireless/ath/ath10k/htt_rx.c
index a39fbf4..62ea9c8 100644
--- a/drivers/net/wireless/ath/ath10k/htt_rx.c
+++ b/drivers/net/wireless/ath/ath10k/htt_rx.c
@@ -1140,7 +1140,7 @@ void ath10k_htt_t2h_msg_handler(struct ath10k *ar, struct sk_buff *skb)
 			break;
 		}
 
-		ath10k_txrx_tx_completed(htt, &tx_done);
+		ath10k_txrx_tx_unref(htt, &tx_done);
 		break;
 	}
 	case HTT_T2H_MSG_TYPE_TX_COMPL_IND: {
@@ -1174,7 +1174,7 @@ void ath10k_htt_t2h_msg_handler(struct ath10k *ar, struct sk_buff *skb)
 		for (i = 0; i < resp->data_tx_completion.num_msdus; i++) {
 			msdu_id = resp->data_tx_completion.msdus[i];
 			tx_done.msdu_id = __le16_to_cpu(msdu_id);
-			ath10k_txrx_tx_completed(htt, &tx_done);
+			ath10k_txrx_tx_unref(htt, &tx_done);
 		}
 		break;
 	}
diff --git a/drivers/net/wireless/ath/ath10k/htt_tx.c b/drivers/net/wireless/ath/ath10k/htt_tx.c
index c4bbf74..06946d2 100644
--- a/drivers/net/wireless/ath/ath10k/htt_tx.c
+++ b/drivers/net/wireless/ath/ath10k/htt_tx.c
@@ -117,7 +117,7 @@ int ath10k_htt_tx_attach(struct ath10k_htt *htt)
 
 static void ath10k_htt_tx_cleanup_pending(struct ath10k_htt *htt)
 {
-	struct sk_buff *txdesc;
+	struct htt_tx_done tx_done = {0};
 	int msdu_id;
 
 	/* No locks needed. Called after communication with the device has
@@ -127,18 +127,13 @@ static void ath10k_htt_tx_cleanup_pending(struct ath10k_htt *htt)
 		if (!test_bit(msdu_id, htt->used_msdu_ids))
 			continue;
 
-		txdesc = htt->pending_tx[msdu_id];
-		if (!txdesc)
-			continue;
-
 		ath10k_dbg(ATH10K_DBG_HTT, "force cleanup msdu_id %hu\n",
 			   msdu_id);
 
-		if (ATH10K_SKB_CB(txdesc)->htt.refcount > 0)
-			ATH10K_SKB_CB(txdesc)->htt.refcount = 1;
+		tx_done.discard = 1;
+		tx_done.msdu_id = msdu_id;
 
-		ATH10K_SKB_CB(txdesc)->htt.discard = true;
-		ath10k_txrx_tx_unref(htt, txdesc);
+		ath10k_txrx_tx_unref(htt, &tx_done);
 	}
 }
 
@@ -152,26 +147,7 @@ void ath10k_htt_tx_detach(struct ath10k_htt *htt)
 
 void ath10k_htt_htc_tx_complete(struct ath10k *ar, struct sk_buff *skb)
 {
-	struct ath10k_skb_cb *skb_cb = ATH10K_SKB_CB(skb);
-	struct ath10k_htt *htt = &ar->htt;
-
-	if (skb_cb->htt.is_conf) {
-		dev_kfree_skb_any(skb);
-		return;
-	}
-
-	if (skb_cb->is_aborted) {
-		skb_cb->htt.discard = true;
-
-		/* if the skbuff is aborted we need to make sure we'll free up
-		 * the tx resources, we can't simply run tx_unref() 2 times
-		 * because if htt tx completion came in earlier we'd access
-		 * unallocated memory */
-		if (skb_cb->htt.refcount > 1)
-			skb_cb->htt.refcount = 1;
-	}
-
-	ath10k_txrx_tx_unref(htt, skb);
+	dev_kfree_skb_any(skb);
 }
 
 int ath10k_htt_h2t_ver_req_msg(struct ath10k_htt *htt)
@@ -192,8 +168,6 @@ int ath10k_htt_h2t_ver_req_msg(struct ath10k_htt *htt)
 	cmd = (struct htt_cmd *)skb->data;
 	cmd->hdr.msg_type = HTT_H2T_MSG_TYPE_VERSION_REQ;
 
-	ATH10K_SKB_CB(skb)->htt.is_conf = true;
-
 	ret = ath10k_htc_send(&htt->ar->htc, htt->eid, skb);
 	if (ret) {
 		dev_kfree_skb_any(skb);
@@ -233,8 +207,6 @@ int ath10k_htt_h2t_stats_req(struct ath10k_htt *htt, u8 mask, u64 cookie)
 	req->cookie_lsb = cpu_to_le32(cookie & 0xffffffff);
 	req->cookie_msb = cpu_to_le32((cookie & 0xffffffff00000000ULL) >> 32);
 
-	ATH10K_SKB_CB(skb)->htt.is_conf = true;
-
 	ret = ath10k_htc_send(&htt->ar->htc, htt->eid, skb);
 	if (ret) {
 		ath10k_warn("failed to send htt type stats request: %d", ret);
@@ -321,8 +293,6 @@ int ath10k_htt_send_rx_ring_cfg_ll(struct ath10k_htt *htt)
 
 #undef desc_offset
 
-	ATH10K_SKB_CB(skb)->htt.is_conf = true;
-
 	ret = ath10k_htc_send(&htt->ar->htc, htt->eid, skb);
 	if (ret) {
 		dev_kfree_skb_any(skb);
@@ -335,7 +305,6 @@ int ath10k_htt_send_rx_ring_cfg_ll(struct ath10k_htt *htt)
 int ath10k_htt_mgmt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 {
 	struct device *dev = htt->ar->dev;
-	struct ath10k_skb_cb *skb_cb;
 	struct sk_buff *txdesc = NULL;
 	struct htt_cmd *cmd;
 	u8 vdev_id = ATH10K_SKB_CB(msdu)->htt.vdev_id;
@@ -364,7 +333,7 @@ int ath10k_htt_mgmt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 		res = msdu_id;
 		goto err;
 	}
-	htt->pending_tx[msdu_id] = txdesc;
+	htt->pending_tx[msdu_id] = msdu;
 	spin_unlock_bh(&htt->tx_lock);
 
 	res = ath10k_skb_map(dev, msdu);
@@ -381,15 +350,6 @@ int ath10k_htt_mgmt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 	memcpy(cmd->mgmt_tx.hdr, msdu->data,
 	       min_t(int, msdu->len, HTT_MGMT_FRM_HDR_DOWNLOAD_LEN));
 
-	/* refcount is decremented by HTC and HTT completions until it reaches
-	 * zero and is freed */
-	skb_cb = ATH10K_SKB_CB(txdesc);
-	skb_cb->htt.is_conf = false;
-	skb_cb->htt.msdu_id = msdu_id;
-	skb_cb->htt.refcount = 2;
-	skb_cb->htt.msdu = msdu;
-	skb_cb->htt.txfrag = NULL;
-
 	res = ath10k_htc_send(&htt->ar->htc, htt->eid, txdesc);
 	if (res)
 		goto err;
@@ -417,7 +377,6 @@ int ath10k_htt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 	struct htt_cmd *cmd;
 	struct htt_data_tx_desc_frag *tx_frags;
 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)msdu->data;
-	struct ath10k_skb_cb *skb_cb;
 	struct sk_buff *txdesc = NULL;
 	struct sk_buff *txfrag = NULL;
 	u8 vdev_id = ATH10K_SKB_CB(msdu)->htt.vdev_id;
@@ -470,7 +429,7 @@ int ath10k_htt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 		res = msdu_id;
 		goto err;
 	}
-	htt->pending_tx[msdu_id] = txdesc;
+	htt->pending_tx[msdu_id] = msdu;
 	spin_unlock_bh(&htt->tx_lock);
 
 	res = ath10k_skb_map(dev, msdu);
@@ -553,15 +512,6 @@ int ath10k_htt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 
 	memcpy(cmd->data_tx.prefetch, msdu->data, prefetch_len);
 
-	/* refcount is decremented by HTC and HTT completions until it reaches
-	 * zero and is freed */
-	skb_cb = ATH10K_SKB_CB(txdesc);
-	skb_cb->htt.is_conf = false;
-	skb_cb->htt.msdu_id = msdu_id;
-	skb_cb->htt.refcount = 2;
-	skb_cb->htt.txfrag = txfrag;
-	skb_cb->htt.msdu = msdu;
-
 	res = ath10k_htc_send(&htt->ar->htc, htt->eid, txdesc);
 	if (res)
 		goto err;
diff --git a/drivers/net/wireless/ath/ath10k/mac.c b/drivers/net/wireless/ath/ath10k/mac.c
index 0d367e4..9112e6d6f 100644
--- a/drivers/net/wireless/ath/ath10k/mac.c
+++ b/drivers/net/wireless/ath/ath10k/mac.c
@@ -1757,8 +1757,6 @@ static void ath10k_tx(struct ieee80211_hw *hw,
 		ath10k_tx_h_seq_no(skb);
 	}
 
-	ATH10K_SKB_CB(skb)->is_mapped = false;
-	ATH10K_SKB_CB(skb)->is_aborted = false;
 	ATH10K_SKB_CB(skb)->htt.is_offchan = false;
 	ATH10K_SKB_CB(skb)->htt.vdev_id = vdev_id;
 	ATH10K_SKB_CB(skb)->htt.tid = tid;
diff --git a/drivers/net/wireless/ath/ath10k/txrx.c b/drivers/net/wireless/ath/ath10k/txrx.c
index f6fed31..15395af 100644
--- a/drivers/net/wireless/ath/ath10k/txrx.c
+++ b/drivers/net/wireless/ath/ath10k/txrx.c
@@ -44,21 +44,25 @@ out:
 	spin_unlock_bh(&ar->data_lock);
 }
 
-void ath10k_txrx_tx_unref(struct ath10k_htt *htt, struct sk_buff *txdesc)
+void ath10k_txrx_tx_unref(struct ath10k_htt *htt,
+			  const struct htt_tx_done *tx_done)
 {
 	struct device *dev = htt->ar->dev;
 	struct ieee80211_tx_info *info;
-	struct sk_buff *txfrag = ATH10K_SKB_CB(txdesc)->htt.txfrag;
-	struct sk_buff *msdu = ATH10K_SKB_CB(txdesc)->htt.msdu;
+	struct sk_buff *msdu, *txfrag;
 	int ret;
 
-	if (ATH10K_SKB_CB(txdesc)->htt.refcount == 0)
-		return;
-
-	ATH10K_SKB_CB(txdesc)->htt.refcount--;
+	ath10k_dbg(ATH10K_DBG_HTT, "htt tx completion msdu_id %u discard %d no_ack %d\n",
+		   tx_done->msdu_id, !!tx_done->discard, !!tx_done->no_ack);
 
-	if (ATH10K_SKB_CB(txdesc)->htt.refcount > 0)
+	if (tx_done->msdu_id >= htt->max_num_pending_tx) {
+		ath10k_warn("warning: msdu_id %d too big, ignoring\n",
+			    tx_done->msdu_id);
 		return;
+	}
+
+	msdu = htt->pending_tx[tx_done->msdu_id];
+	txfrag = ATH10K_SKB_CB(msdu)->htt.txfrag;
 
 	if (txfrag) {
 		ret = ath10k_skb_unmap(dev, txfrag);
@@ -76,7 +80,7 @@ void ath10k_txrx_tx_unref(struct ath10k_htt *htt, struct sk_buff *txdesc)
 
 	info = IEEE80211_SKB_CB(msdu);
 
-	if (ATH10K_SKB_CB(txdesc)->htt.discard) {
+	if (tx_done->discard) {
 		ieee80211_free_txskb(htt->ar->hw, msdu);
 		goto exit;
 	}
@@ -84,7 +88,7 @@ void ath10k_txrx_tx_unref(struct ath10k_htt *htt, struct sk_buff *txdesc)
 	if (!(info->flags & IEEE80211_TX_CTL_NO_ACK))
 		info->flags |= IEEE80211_TX_STAT_ACK;
 
-	if (ATH10K_SKB_CB(txdesc)->htt.no_ack)
+	if (tx_done->no_ack)
 		info->flags &= ~IEEE80211_TX_STAT_ACK;
 
 	ieee80211_tx_status(htt->ar->hw, msdu);
@@ -92,36 +96,12 @@ void ath10k_txrx_tx_unref(struct ath10k_htt *htt, struct sk_buff *txdesc)
 
 exit:
 	spin_lock_bh(&htt->tx_lock);
-	htt->pending_tx[ATH10K_SKB_CB(txdesc)->htt.msdu_id] = NULL;
-	ath10k_htt_tx_free_msdu_id(htt, ATH10K_SKB_CB(txdesc)->htt.msdu_id);
+	htt->pending_tx[tx_done->msdu_id] = NULL;
+	ath10k_htt_tx_free_msdu_id(htt, tx_done->msdu_id);
 	__ath10k_htt_tx_dec_pending(htt);
 	if (htt->num_pending_tx == 0)
 		wake_up(&htt->empty_tx_wq);
 	spin_unlock_bh(&htt->tx_lock);
-
-	dev_kfree_skb_any(txdesc);
-}
-
-void ath10k_txrx_tx_completed(struct ath10k_htt *htt,
-			      const struct htt_tx_done *tx_done)
-{
-	struct sk_buff *txdesc;
-
-	ath10k_dbg(ATH10K_DBG_HTT, "htt tx completion msdu_id %u discard %d no_ack %d\n",
-		   tx_done->msdu_id, !!tx_done->discard, !!tx_done->no_ack);
-
-	if (tx_done->msdu_id >= htt->max_num_pending_tx) {
-		ath10k_warn("warning: msdu_id %d too big, ignoring\n",
-			    tx_done->msdu_id);
-		return;
-	}
-
-	txdesc = htt->pending_tx[tx_done->msdu_id];
-
-	ATH10K_SKB_CB(txdesc)->htt.discard = tx_done->discard;
-	ATH10K_SKB_CB(txdesc)->htt.no_ack = tx_done->no_ack;
-
-	ath10k_txrx_tx_unref(htt, txdesc);
 }
 
 static const u8 rx_legacy_rate_idx[] = {
diff --git a/drivers/net/wireless/ath/ath10k/txrx.h b/drivers/net/wireless/ath/ath10k/txrx.h
index e78632a..356dc9c 100644
--- a/drivers/net/wireless/ath/ath10k/txrx.h
+++ b/drivers/net/wireless/ath/ath10k/txrx.h
@@ -19,9 +19,8 @@
 
 #include "htt.h"
 
-void ath10k_txrx_tx_unref(struct ath10k_htt *htt, struct sk_buff *txdesc);
-void ath10k_txrx_tx_completed(struct ath10k_htt *htt,
-			      const struct htt_tx_done *tx_done);
+void ath10k_txrx_tx_unref(struct ath10k_htt *htt,
+			  const struct htt_tx_done *tx_done);
 void ath10k_process_rx(struct ath10k *ar, struct htt_rx_info *info);
 
 struct ath10k_peer *ath10k_peer_find(struct ath10k *ar, int vdev_id,
-- 
1.7.9.5


_______________________________________________
ath10k mailing list
ath10k@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/ath10k

[PATCH 3/5] ath10k: decouple HTT TX completions

[Michal Kazior]

Until now the all MSDU transfer related structures
were freed when all resources were unreferenced.

Now HTC transfer is freed independently and HTT
transfer is so too.

This yields a way more simpler ath10k_skb_cb and
should possibly enable parallel pipe processing
(which is now serialized in
ath10k_pci_process_ce routine).

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
---
 drivers/net/wireless/ath/ath10k/core.h   |    8 ----
 drivers/net/wireless/ath/ath10k/htt_rx.c |    4 +-
 drivers/net/wireless/ath/ath10k/htt_tx.c |   64 ++++--------------------------
 drivers/net/wireless/ath/ath10k/mac.c    |    2 -
 drivers/net/wireless/ath/ath10k/txrx.c   |   52 ++++++++----------------
 drivers/net/wireless/ath/ath10k/txrx.h   |    5 +--
 6 files changed, 27 insertions(+), 108 deletions(-)

diff --git a/drivers/net/wireless/ath/ath10k/core.h b/drivers/net/wireless/ath/ath10k/core.h
index fcf94ee..4563f80 100644
--- a/drivers/net/wireless/ath/ath10k/core.h
+++ b/drivers/net/wireless/ath/ath10k/core.h
@@ -52,18 +52,10 @@ struct ath10k_skb_cb {
 
 	struct {
 		u8 vdev_id;
-		u16 msdu_id;
 		u8 tid;
 		bool is_offchan;
-		bool is_conf;
-		bool discard;
-		bool no_ack;
-		u8 refcount;
 		struct sk_buff *txfrag;
-		struct sk_buff *msdu;
 	} __packed htt;
-
-	/* 4 bytes left on 64bit arch */
 } __packed;
 
 static inline struct ath10k_skb_cb *ATH10K_SKB_CB(struct sk_buff *skb)
diff --git a/drivers/net/wireless/ath/ath10k/htt_rx.c b/drivers/net/wireless/ath/ath10k/htt_rx.c
index a39fbf4..62ea9c8 100644
--- a/drivers/net/wireless/ath/ath10k/htt_rx.c
+++ b/drivers/net/wireless/ath/ath10k/htt_rx.c
@@ -1140,7 +1140,7 @@ void ath10k_htt_t2h_msg_handler(struct ath10k *ar, struct sk_buff *skb)
 			break;
 		}
 
-		ath10k_txrx_tx_completed(htt, &tx_done);
+		ath10k_txrx_tx_unref(htt, &tx_done);
 		break;
 	}
 	case HTT_T2H_MSG_TYPE_TX_COMPL_IND: {
@@ -1174,7 +1174,7 @@ void ath10k_htt_t2h_msg_handler(struct ath10k *ar, struct sk_buff *skb)
 		for (i = 0; i < resp->data_tx_completion.num_msdus; i++) {
 			msdu_id = resp->data_tx_completion.msdus[i];
 			tx_done.msdu_id = __le16_to_cpu(msdu_id);
-			ath10k_txrx_tx_completed(htt, &tx_done);
+			ath10k_txrx_tx_unref(htt, &tx_done);
 		}
 		break;
 	}
diff --git a/drivers/net/wireless/ath/ath10k/htt_tx.c b/drivers/net/wireless/ath/ath10k/htt_tx.c
index c4bbf74..06946d2 100644
--- a/drivers/net/wireless/ath/ath10k/htt_tx.c
+++ b/drivers/net/wireless/ath/ath10k/htt_tx.c
@@ -117,7 +117,7 @@ int ath10k_htt_tx_attach(struct ath10k_htt *htt)
 
 static void ath10k_htt_tx_cleanup_pending(struct ath10k_htt *htt)
 {
-	struct sk_buff *txdesc;
+	struct htt_tx_done tx_done = {0};
 	int msdu_id;
 
 	/* No locks needed. Called after communication with the device has
@@ -127,18 +127,13 @@ static void ath10k_htt_tx_cleanup_pending(struct ath10k_htt *htt)
 		if (!test_bit(msdu_id, htt->used_msdu_ids))
 			continue;
 
-		txdesc = htt->pending_tx[msdu_id];
-		if (!txdesc)
-			continue;
-
 		ath10k_dbg(ATH10K_DBG_HTT, "force cleanup msdu_id %hu\n",
 			   msdu_id);
 
-		if (ATH10K_SKB_CB(txdesc)->htt.refcount > 0)
-			ATH10K_SKB_CB(txdesc)->htt.refcount = 1;
+		tx_done.discard = 1;
+		tx_done.msdu_id = msdu_id;
 
-		ATH10K_SKB_CB(txdesc)->htt.discard = true;
-		ath10k_txrx_tx_unref(htt, txdesc);
+		ath10k_txrx_tx_unref(htt, &tx_done);
 	}
 }
 
@@ -152,26 +147,7 @@ void ath10k_htt_tx_detach(struct ath10k_htt *htt)
 
 void ath10k_htt_htc_tx_complete(struct ath10k *ar, struct sk_buff *skb)
 {
-	struct ath10k_skb_cb *skb_cb = ATH10K_SKB_CB(skb);
-	struct ath10k_htt *htt = &ar->htt;
-
-	if (skb_cb->htt.is_conf) {
-		dev_kfree_skb_any(skb);
-		return;
-	}
-
-	if (skb_cb->is_aborted) {
-		skb_cb->htt.discard = true;
-
-		/* if the skbuff is aborted we need to make sure we'll free up
-		 * the tx resources, we can't simply run tx_unref() 2 times
-		 * because if htt tx completion came in earlier we'd access
-		 * unallocated memory */
-		if (skb_cb->htt.refcount > 1)
-			skb_cb->htt.refcount = 1;
-	}
-
-	ath10k_txrx_tx_unref(htt, skb);
+	dev_kfree_skb_any(skb);
 }
 
 int ath10k_htt_h2t_ver_req_msg(struct ath10k_htt *htt)
@@ -192,8 +168,6 @@ int ath10k_htt_h2t_ver_req_msg(struct ath10k_htt *htt)
 	cmd = (struct htt_cmd *)skb->data;
 	cmd->hdr.msg_type = HTT_H2T_MSG_TYPE_VERSION_REQ;
 
-	ATH10K_SKB_CB(skb)->htt.is_conf = true;
-
 	ret = ath10k_htc_send(&htt->ar->htc, htt->eid, skb);
 	if (ret) {
 		dev_kfree_skb_any(skb);
@@ -233,8 +207,6 @@ int ath10k_htt_h2t_stats_req(struct ath10k_htt *htt, u8 mask, u64 cookie)
 	req->cookie_lsb = cpu_to_le32(cookie & 0xffffffff);
 	req->cookie_msb = cpu_to_le32((cookie & 0xffffffff00000000ULL) >> 32);
 
-	ATH10K_SKB_CB(skb)->htt.is_conf = true;
-
 	ret = ath10k_htc_send(&htt->ar->htc, htt->eid, skb);
 	if (ret) {
 		ath10k_warn("failed to send htt type stats request: %d", ret);
@@ -321,8 +293,6 @@ int ath10k_htt_send_rx_ring_cfg_ll(struct ath10k_htt *htt)
 
 #undef desc_offset
 
-	ATH10K_SKB_CB(skb)->htt.is_conf = true;
-
 	ret = ath10k_htc_send(&htt->ar->htc, htt->eid, skb);
 	if (ret) {
 		dev_kfree_skb_any(skb);
@@ -335,7 +305,6 @@ int ath10k_htt_send_rx_ring_cfg_ll(struct ath10k_htt *htt)
 int ath10k_htt_mgmt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 {
 	struct device *dev = htt->ar->dev;
-	struct ath10k_skb_cb *skb_cb;
 	struct sk_buff *txdesc = NULL;
 	struct htt_cmd *cmd;
 	u8 vdev_id = ATH10K_SKB_CB(msdu)->htt.vdev_id;
@@ -364,7 +333,7 @@ int ath10k_htt_mgmt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 		res = msdu_id;
 		goto err;
 	}
-	htt->pending_tx[msdu_id] = txdesc;
+	htt->pending_tx[msdu_id] = msdu;
 	spin_unlock_bh(&htt->tx_lock);
 
 	res = ath10k_skb_map(dev, msdu);
@@ -381,15 +350,6 @@ int ath10k_htt_mgmt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 	memcpy(cmd->mgmt_tx.hdr, msdu->data,
 	       min_t(int, msdu->len, HTT_MGMT_FRM_HDR_DOWNLOAD_LEN));
 
-	/* refcount is decremented by HTC and HTT completions until it reaches
-	 * zero and is freed */
-	skb_cb = ATH10K_SKB_CB(txdesc);
-	skb_cb->htt.is_conf = false;
-	skb_cb->htt.msdu_id = msdu_id;
-	skb_cb->htt.refcount = 2;
-	skb_cb->htt.msdu = msdu;
-	skb_cb->htt.txfrag = NULL;
-
 	res = ath10k_htc_send(&htt->ar->htc, htt->eid, txdesc);
 	if (res)
 		goto err;
@@ -417,7 +377,6 @@ int ath10k_htt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 	struct htt_cmd *cmd;
 	struct htt_data_tx_desc_frag *tx_frags;
 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)msdu->data;
-	struct ath10k_skb_cb *skb_cb;
 	struct sk_buff *txdesc = NULL;
 	struct sk_buff *txfrag = NULL;
 	u8 vdev_id = ATH10K_SKB_CB(msdu)->htt.vdev_id;
@@ -470,7 +429,7 @@ int ath10k_htt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 		res = msdu_id;
 		goto err;
 	}
-	htt->pending_tx[msdu_id] = txdesc;
+	htt->pending_tx[msdu_id] = msdu;
 	spin_unlock_bh(&htt->tx_lock);
 
 	res = ath10k_skb_map(dev, msdu);
@@ -553,15 +512,6 @@ int ath10k_htt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 
 	memcpy(cmd->data_tx.prefetch, msdu->data, prefetch_len);
 
-	/* refcount is decremented by HTC and HTT completions until it reaches
-	 * zero and is freed */
-	skb_cb = ATH10K_SKB_CB(txdesc);
-	skb_cb->htt.is_conf = false;
-	skb_cb->htt.msdu_id = msdu_id;
-	skb_cb->htt.refcount = 2;
-	skb_cb->htt.txfrag = txfrag;
-	skb_cb->htt.msdu = msdu;
-
 	res = ath10k_htc_send(&htt->ar->htc, htt->eid, txdesc);
 	if (res)
 		goto err;
diff --git a/drivers/net/wireless/ath/ath10k/mac.c b/drivers/net/wireless/ath/ath10k/mac.c
index 0d367e4..9112e6d6f 100644
--- a/drivers/net/wireless/ath/ath10k/mac.c
+++ b/drivers/net/wireless/ath/ath10k/mac.c
@@ -1757,8 +1757,6 @@ static void ath10k_tx(struct ieee80211_hw *hw,
 		ath10k_tx_h_seq_no(skb);
 	}
 
-	ATH10K_SKB_CB(skb)->is_mapped = false;
-	ATH10K_SKB_CB(skb)->is_aborted = false;
 	ATH10K_SKB_CB(skb)->htt.is_offchan = false;
 	ATH10K_SKB_CB(skb)->htt.vdev_id = vdev_id;
 	ATH10K_SKB_CB(skb)->htt.tid = tid;
diff --git a/drivers/net/wireless/ath/ath10k/txrx.c b/drivers/net/wireless/ath/ath10k/txrx.c
index f6fed31..15395af 100644
--- a/drivers/net/wireless/ath/ath10k/txrx.c
+++ b/drivers/net/wireless/ath/ath10k/txrx.c
@@ -44,21 +44,25 @@ out:
 	spin_unlock_bh(&ar->data_lock);
 }
 
-void ath10k_txrx_tx_unref(struct ath10k_htt *htt, struct sk_buff *txdesc)
+void ath10k_txrx_tx_unref(struct ath10k_htt *htt,
+			  const struct htt_tx_done *tx_done)
 {
 	struct device *dev = htt->ar->dev;
 	struct ieee80211_tx_info *info;
-	struct sk_buff *txfrag = ATH10K_SKB_CB(txdesc)->htt.txfrag;
-	struct sk_buff *msdu = ATH10K_SKB_CB(txdesc)->htt.msdu;
+	struct sk_buff *msdu, *txfrag;
 	int ret;
 
-	if (ATH10K_SKB_CB(txdesc)->htt.refcount == 0)
-		return;
-
-	ATH10K_SKB_CB(txdesc)->htt.refcount--;
+	ath10k_dbg(ATH10K_DBG_HTT, "htt tx completion msdu_id %u discard %d no_ack %d\n",
+		   tx_done->msdu_id, !!tx_done->discard, !!tx_done->no_ack);
 
-	if (ATH10K_SKB_CB(txdesc)->htt.refcount > 0)
+	if (tx_done->msdu_id >= htt->max_num_pending_tx) {
+		ath10k_warn("warning: msdu_id %d too big, ignoring\n",
+			    tx_done->msdu_id);
 		return;
+	}
+
+	msdu = htt->pending_tx[tx_done->msdu_id];
+	txfrag = ATH10K_SKB_CB(msdu)->htt.txfrag;
 
 	if (txfrag) {
 		ret = ath10k_skb_unmap(dev, txfrag);
@@ -76,7 +80,7 @@ void ath10k_txrx_tx_unref(struct ath10k_htt *htt, struct sk_buff *txdesc)
 
 	info = IEEE80211_SKB_CB(msdu);
 
-	if (ATH10K_SKB_CB(txdesc)->htt.discard) {
+	if (tx_done->discard) {
 		ieee80211_free_txskb(htt->ar->hw, msdu);
 		goto exit;
 	}
@@ -84,7 +88,7 @@ void ath10k_txrx_tx_unref(struct ath10k_htt *htt, struct sk_buff *txdesc)
 	if (!(info->flags & IEEE80211_TX_CTL_NO_ACK))
 		info->flags |= IEEE80211_TX_STAT_ACK;
 
-	if (ATH10K_SKB_CB(txdesc)->htt.no_ack)
+	if (tx_done->no_ack)
 		info->flags &= ~IEEE80211_TX_STAT_ACK;
 
 	ieee80211_tx_status(htt->ar->hw, msdu);
@@ -92,36 +96,12 @@ void ath10k_txrx_tx_unref(struct ath10k_htt *htt, struct sk_buff *txdesc)
 
 exit:
 	spin_lock_bh(&htt->tx_lock);
-	htt->pending_tx[ATH10K_SKB_CB(txdesc)->htt.msdu_id] = NULL;
-	ath10k_htt_tx_free_msdu_id(htt, ATH10K_SKB_CB(txdesc)->htt.msdu_id);
+	htt->pending_tx[tx_done->msdu_id] = NULL;
+	ath10k_htt_tx_free_msdu_id(htt, tx_done->msdu_id);
 	__ath10k_htt_tx_dec_pending(htt);
 	if (htt->num_pending_tx == 0)
 		wake_up(&htt->empty_tx_wq);
 	spin_unlock_bh(&htt->tx_lock);
-
-	dev_kfree_skb_any(txdesc);
-}
-
-void ath10k_txrx_tx_completed(struct ath10k_htt *htt,
-			      const struct htt_tx_done *tx_done)
-{
-	struct sk_buff *txdesc;
-
-	ath10k_dbg(ATH10K_DBG_HTT, "htt tx completion msdu_id %u discard %d no_ack %d\n",
-		   tx_done->msdu_id, !!tx_done->discard, !!tx_done->no_ack);
-
-	if (tx_done->msdu_id >= htt->max_num_pending_tx) {
-		ath10k_warn("warning: msdu_id %d too big, ignoring\n",
-			    tx_done->msdu_id);
-		return;
-	}
-
-	txdesc = htt->pending_tx[tx_done->msdu_id];
-
-	ATH10K_SKB_CB(txdesc)->htt.discard = tx_done->discard;
-	ATH10K_SKB_CB(txdesc)->htt.no_ack = tx_done->no_ack;
-
-	ath10k_txrx_tx_unref(htt, txdesc);
 }
 
 static const u8 rx_legacy_rate_idx[] = {
diff --git a/drivers/net/wireless/ath/ath10k/txrx.h b/drivers/net/wireless/ath/ath10k/txrx.h
index e78632a..356dc9c 100644
--- a/drivers/net/wireless/ath/ath10k/txrx.h
+++ b/drivers/net/wireless/ath/ath10k/txrx.h
@@ -19,9 +19,8 @@
 
 #include "htt.h"
 
-void ath10k_txrx_tx_unref(struct ath10k_htt *htt, struct sk_buff *txdesc);
-void ath10k_txrx_tx_completed(struct ath10k_htt *htt,
-			      const struct htt_tx_done *tx_done);
+void ath10k_txrx_tx_unref(struct ath10k_htt *htt,
+			  const struct htt_tx_done *tx_done);
 void ath10k_process_rx(struct ath10k *ar, struct htt_rx_info *info);
 
 struct ath10k_peer *ath10k_peer_find(struct ath10k *ar, int vdev_id,
-- 
1.7.9.5

[PATCH 4/5] ath10k: cleanup HTT TX functions

[Michal Kazior]

Use a saner goto scheme for failure handling. Also
group operations more sensibly.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
---
 drivers/net/wireless/ath/ath10k/htt_tx.c |  141 ++++++++++++++----------------
 1 file changed, 68 insertions(+), 73 deletions(-)

diff --git a/drivers/net/wireless/ath/ath10k/htt_tx.c b/drivers/net/wireless/ath/ath10k/htt_tx.c
index 06946d2..5e738d8 100644
--- a/drivers/net/wireless/ath/ath10k/htt_tx.c
+++ b/drivers/net/wireless/ath/ath10k/htt_tx.c
@@ -315,30 +315,30 @@ int ath10k_htt_mgmt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 
 	res = ath10k_htt_tx_inc_pending(htt);
 	if (res)
-		return res;
+		goto err;
 
 	len += sizeof(cmd->hdr);
 	len += sizeof(cmd->mgmt_tx);
 
-	txdesc = ath10k_htc_alloc_skb(len);
-	if (!txdesc) {
-		res = -ENOMEM;
-		goto err;
-	}
-
 	spin_lock_bh(&htt->tx_lock);
-	msdu_id = ath10k_htt_tx_alloc_msdu_id(htt);
-	if (msdu_id < 0) {
+	res = ath10k_htt_tx_alloc_msdu_id(htt);
+	if (res < 0) {
 		spin_unlock_bh(&htt->tx_lock);
-		res = msdu_id;
-		goto err;
+		goto err_tx_dec;
 	}
+	msdu_id = res;
 	htt->pending_tx[msdu_id] = msdu;
 	spin_unlock_bh(&htt->tx_lock);
 
+	txdesc = ath10k_htc_alloc_skb(len);
+	if (!txdesc) {
+		res = -ENOMEM;
+		goto err_free_msdu_id;
+	}
+
 	res = ath10k_skb_map(dev, msdu);
 	if (res)
-		goto err;
+		goto err_free_txdesc;
 
 	skb_put(txdesc, len);
 	cmd = (struct htt_cmd *)txdesc->data;
@@ -352,22 +352,22 @@ int ath10k_htt_mgmt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 
 	res = ath10k_htc_send(&htt->ar->htc, htt->eid, txdesc);
 	if (res)
-		goto err;
+		goto err_unmap_msdu;
 
 	return 0;
 
-err:
+err_unmap_msdu:
 	ath10k_skb_unmap(dev, msdu);
-
-	if (txdesc)
-		dev_kfree_skb_any(txdesc);
-	if (msdu_id >= 0) {
-		spin_lock_bh(&htt->tx_lock);
-		htt->pending_tx[msdu_id] = NULL;
-		ath10k_htt_tx_free_msdu_id(htt, msdu_id);
-		spin_unlock_bh(&htt->tx_lock);
-	}
+err_free_txdesc:
+	dev_kfree_skb_any(txdesc);
+err_free_msdu_id:
+	spin_lock_bh(&htt->tx_lock);
+	htt->pending_tx[msdu_id] = NULL;
+	ath10k_htt_tx_free_msdu_id(htt, msdu_id);
+	spin_unlock_bh(&htt->tx_lock);
+err_tx_dec:
 	ath10k_htt_tx_dec_pending(htt);
+err:
 	return res;
 }
 
@@ -379,6 +379,7 @@ int ath10k_htt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)msdu->data;
 	struct sk_buff *txdesc = NULL;
 	struct sk_buff *txfrag = NULL;
+	bool use_frags;
 	u8 vdev_id = ATH10K_SKB_CB(msdu)->htt.vdev_id;
 	u8 tid;
 	int prefetch_len, desc_len, frag_len;
@@ -390,7 +391,17 @@ int ath10k_htt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 
 	res = ath10k_htt_tx_inc_pending(htt);
 	if (res)
-		return res;
+		goto err;
+
+	spin_lock_bh(&htt->tx_lock);
+	res = ath10k_htt_tx_alloc_msdu_id(htt);
+	if (res < 0) {
+		spin_unlock_bh(&htt->tx_lock);
+		goto err_tx_dec;
+	}
+	msdu_id = res;
+	htt->pending_tx[msdu_id] = msdu;
+	spin_unlock_bh(&htt->tx_lock);
 
 	prefetch_len = min(htt->prefetch_len, msdu->len);
 	prefetch_len = roundup(prefetch_len, 4);
@@ -401,46 +412,34 @@ int ath10k_htt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 	txdesc = ath10k_htc_alloc_skb(desc_len);
 	if (!txdesc) {
 		res = -ENOMEM;
-		goto err;
+		goto err_free_msdu_id;
 	}
 
 	/* Since HTT 3.0 there is no separate mgmt tx command. However in case
 	 * of mgmt tx using TX_FRM there is not tx fragment list. Instead of tx
 	 * fragment list host driver specifies directly frame pointer. */
-	if (htt->target_version_major < 3 ||
-	    !ieee80211_is_mgmt(hdr->frame_control)) {
+	use_frags = htt->target_version_major < 3 ||
+		    !ieee80211_is_mgmt(hdr->frame_control);
+
+	if (use_frags) {
 		txfrag = dev_alloc_skb(frag_len);
 		if (!txfrag) {
 			res = -ENOMEM;
-			goto err;
+			goto err_free_txdesc;
 		}
 	}
 
 	if (!IS_ALIGNED((unsigned long)txdesc->data, 4)) {
 		ath10k_warn("htt alignment check failed. dropping packet.\n");
 		res = -EIO;
-		goto err;
+		goto err_free_txfrag;
 	}
 
-	spin_lock_bh(&htt->tx_lock);
-	msdu_id = ath10k_htt_tx_alloc_msdu_id(htt);
-	if (msdu_id < 0) {
-		spin_unlock_bh(&htt->tx_lock);
-		res = msdu_id;
-		goto err;
-	}
-	htt->pending_tx[msdu_id] = msdu;
-	spin_unlock_bh(&htt->tx_lock);
-
 	res = ath10k_skb_map(dev, msdu);
 	if (res)
-		goto err;
+		goto err_free_txfrag;
 
-	/* Since HTT 3.0 there is no separate mgmt tx command. However in case
-	 * of mgmt tx using TX_FRM there is not tx fragment list. Instead of tx
-	 * fragment list host driver specifies directly frame pointer. */
-	if (htt->target_version_major < 3 ||
-	    !ieee80211_is_mgmt(hdr->frame_control)) {
+	if (use_frags) {
 		/* tx fragment list must be terminated with zero-entry */
 		skb_put(txfrag, frag_len);
 		tx_frags = (struct htt_data_tx_desc_frag *)txfrag->data;
@@ -451,7 +450,7 @@ int ath10k_htt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 
 		res = ath10k_skb_map(dev, txfrag);
 		if (res)
-			goto err;
+			goto err_unmap_msdu;
 
 		ath10k_dbg(ATH10K_DBG_HTT, "txfrag 0x%llx\n",
 			   (unsigned long long) ATH10K_SKB_CB(txfrag)->paddr);
@@ -476,15 +475,11 @@ int ath10k_htt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 		flags0 |= HTT_DATA_TX_DESC_FLAGS0_NO_ENCRYPT;
 	flags0 |= HTT_DATA_TX_DESC_FLAGS0_MAC_HDR_PRESENT;
 
-	/* Since HTT 3.0 there is no separate mgmt tx command. However in case
-	 * of mgmt tx using TX_FRM there is not tx fragment list. Instead of tx
-	 * fragment list host driver specifies directly frame pointer. */
-	if (htt->target_version_major >= 3 &&
-	    ieee80211_is_mgmt(hdr->frame_control))
-		flags0 |= SM(ATH10K_HW_TXRX_MGMT,
+	if (use_frags)
+		flags0 |= SM(ATH10K_HW_TXRX_NATIVE_WIFI,
 			     HTT_DATA_TX_DESC_FLAGS0_PKT_TYPE);
 	else
-		flags0 |= SM(ATH10K_HW_TXRX_NATIVE_WIFI,
+		flags0 |= SM(ATH10K_HW_TXRX_MGMT,
 			     HTT_DATA_TX_DESC_FLAGS0_PKT_TYPE);
 
 	flags1  = 0;
@@ -493,14 +488,10 @@ int ath10k_htt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 	flags1 |= HTT_DATA_TX_DESC_FLAGS1_CKSUM_L3_OFFLOAD;
 	flags1 |= HTT_DATA_TX_DESC_FLAGS1_CKSUM_L4_OFFLOAD;
 
-	/* Since HTT 3.0 there is no separate mgmt tx command. However in case
-	 * of mgmt tx using TX_FRM there is not tx fragment list. Instead of tx
-	 * fragment list host driver specifies directly frame pointer. */
-	if (htt->target_version_major >= 3 &&
-	    ieee80211_is_mgmt(hdr->frame_control))
-		frags_paddr = ATH10K_SKB_CB(msdu)->paddr;
-	else
+	if (use_frags)
 		frags_paddr = ATH10K_SKB_CB(txfrag)->paddr;
+	else
+		frags_paddr = ATH10K_SKB_CB(msdu)->paddr;
 
 	cmd->hdr.msg_type        = HTT_H2T_MSG_TYPE_TX_FRM;
 	cmd->data_tx.flags0      = flags0;
@@ -514,23 +505,27 @@ int ath10k_htt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 
 	res = ath10k_htc_send(&htt->ar->htc, htt->eid, txdesc);
 	if (res)
-		goto err;
+		goto err_restore;
 
 	return 0;
-err:
-	if (txfrag)
+
+err_restore:
+	if (use_frags)
 		ath10k_skb_unmap(dev, txfrag);
-	if (txdesc)
-		dev_kfree_skb_any(txdesc);
-	if (txfrag)
+err_unmap_msdu:
+	ath10k_skb_unmap(dev, msdu);
+err_free_txfrag:
+	if (use_frags)
 		dev_kfree_skb_any(txfrag);
-	if (msdu_id >= 0) {
-		spin_lock_bh(&htt->tx_lock);
-		htt->pending_tx[msdu_id] = NULL;
-		ath10k_htt_tx_free_msdu_id(htt, msdu_id);
-		spin_unlock_bh(&htt->tx_lock);
-	}
+err_free_txdesc:
+	dev_kfree_skb_any(txdesc);
+err_free_msdu_id:
+	spin_lock_bh(&htt->tx_lock);
+	htt->pending_tx[msdu_id] = NULL;
+	ath10k_htt_tx_free_msdu_id(htt, msdu_id);
+	spin_unlock_bh(&htt->tx_lock);
+err_tx_dec:
 	ath10k_htt_tx_dec_pending(htt);
-	ath10k_skb_unmap(dev, msdu);
+err:
 	return res;
 }
-- 
1.7.9.5


_______________________________________________
ath10k mailing list
ath10k@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/ath10k

[PATCH 4/5] ath10k: cleanup HTT TX functions

[Michal Kazior]

Use a saner goto scheme for failure handling. Also
group operations more sensibly.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
---
 drivers/net/wireless/ath/ath10k/htt_tx.c |  141 ++++++++++++++----------------
 1 file changed, 68 insertions(+), 73 deletions(-)

diff --git a/drivers/net/wireless/ath/ath10k/htt_tx.c b/drivers/net/wireless/ath/ath10k/htt_tx.c
index 06946d2..5e738d8 100644
--- a/drivers/net/wireless/ath/ath10k/htt_tx.c
+++ b/drivers/net/wireless/ath/ath10k/htt_tx.c
@@ -315,30 +315,30 @@ int ath10k_htt_mgmt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 
 	res = ath10k_htt_tx_inc_pending(htt);
 	if (res)
-		return res;
+		goto err;
 
 	len += sizeof(cmd->hdr);
 	len += sizeof(cmd->mgmt_tx);
 
-	txdesc = ath10k_htc_alloc_skb(len);
-	if (!txdesc) {
-		res = -ENOMEM;
-		goto err;
-	}
-
 	spin_lock_bh(&htt->tx_lock);
-	msdu_id = ath10k_htt_tx_alloc_msdu_id(htt);
-	if (msdu_id < 0) {
+	res = ath10k_htt_tx_alloc_msdu_id(htt);
+	if (res < 0) {
 		spin_unlock_bh(&htt->tx_lock);
-		res = msdu_id;
-		goto err;
+		goto err_tx_dec;
 	}
+	msdu_id = res;
 	htt->pending_tx[msdu_id] = msdu;
 	spin_unlock_bh(&htt->tx_lock);
 
+	txdesc = ath10k_htc_alloc_skb(len);
+	if (!txdesc) {
+		res = -ENOMEM;
+		goto err_free_msdu_id;
+	}
+
 	res = ath10k_skb_map(dev, msdu);
 	if (res)
-		goto err;
+		goto err_free_txdesc;
 
 	skb_put(txdesc, len);
 	cmd = (struct htt_cmd *)txdesc->data;
@@ -352,22 +352,22 @@ int ath10k_htt_mgmt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 
 	res = ath10k_htc_send(&htt->ar->htc, htt->eid, txdesc);
 	if (res)
-		goto err;
+		goto err_unmap_msdu;
 
 	return 0;
 
-err:
+err_unmap_msdu:
 	ath10k_skb_unmap(dev, msdu);
-
-	if (txdesc)
-		dev_kfree_skb_any(txdesc);
-	if (msdu_id >= 0) {
-		spin_lock_bh(&htt->tx_lock);
-		htt->pending_tx[msdu_id] = NULL;
-		ath10k_htt_tx_free_msdu_id(htt, msdu_id);
-		spin_unlock_bh(&htt->tx_lock);
-	}
+err_free_txdesc:
+	dev_kfree_skb_any(txdesc);
+err_free_msdu_id:
+	spin_lock_bh(&htt->tx_lock);
+	htt->pending_tx[msdu_id] = NULL;
+	ath10k_htt_tx_free_msdu_id(htt, msdu_id);
+	spin_unlock_bh(&htt->tx_lock);
+err_tx_dec:
 	ath10k_htt_tx_dec_pending(htt);
+err:
 	return res;
 }
 
@@ -379,6 +379,7 @@ int ath10k_htt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)msdu->data;
 	struct sk_buff *txdesc = NULL;
 	struct sk_buff *txfrag = NULL;
+	bool use_frags;
 	u8 vdev_id = ATH10K_SKB_CB(msdu)->htt.vdev_id;
 	u8 tid;
 	int prefetch_len, desc_len, frag_len;
@@ -390,7 +391,17 @@ int ath10k_htt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 
 	res = ath10k_htt_tx_inc_pending(htt);
 	if (res)
-		return res;
+		goto err;
+
+	spin_lock_bh(&htt->tx_lock);
+	res = ath10k_htt_tx_alloc_msdu_id(htt);
+	if (res < 0) {
+		spin_unlock_bh(&htt->tx_lock);
+		goto err_tx_dec;
+	}
+	msdu_id = res;
+	htt->pending_tx[msdu_id] = msdu;
+	spin_unlock_bh(&htt->tx_lock);
 
 	prefetch_len = min(htt->prefetch_len, msdu->len);
 	prefetch_len = roundup(prefetch_len, 4);
@@ -401,46 +412,34 @@ int ath10k_htt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 	txdesc = ath10k_htc_alloc_skb(desc_len);
 	if (!txdesc) {
 		res = -ENOMEM;
-		goto err;
+		goto err_free_msdu_id;
 	}
 
 	/* Since HTT 3.0 there is no separate mgmt tx command. However in case
 	 * of mgmt tx using TX_FRM there is not tx fragment list. Instead of tx
 	 * fragment list host driver specifies directly frame pointer. */
-	if (htt->target_version_major < 3 ||
-	    !ieee80211_is_mgmt(hdr->frame_control)) {
+	use_frags = htt->target_version_major < 3 ||
+		    !ieee80211_is_mgmt(hdr->frame_control);
+
+	if (use_frags) {
 		txfrag = dev_alloc_skb(frag_len);
 		if (!txfrag) {
 			res = -ENOMEM;
-			goto err;
+			goto err_free_txdesc;
 		}
 	}
 
 	if (!IS_ALIGNED((unsigned long)txdesc->data, 4)) {
 		ath10k_warn("htt alignment check failed. dropping packet.\n");
 		res = -EIO;
-		goto err;
+		goto err_free_txfrag;
 	}
 
-	spin_lock_bh(&htt->tx_lock);
-	msdu_id = ath10k_htt_tx_alloc_msdu_id(htt);
-	if (msdu_id < 0) {
-		spin_unlock_bh(&htt->tx_lock);
-		res = msdu_id;
-		goto err;
-	}
-	htt->pending_tx[msdu_id] = msdu;
-	spin_unlock_bh(&htt->tx_lock);
-
 	res = ath10k_skb_map(dev, msdu);
 	if (res)
-		goto err;
+		goto err_free_txfrag;
 
-	/* Since HTT 3.0 there is no separate mgmt tx command. However in case
-	 * of mgmt tx using TX_FRM there is not tx fragment list. Instead of tx
-	 * fragment list host driver specifies directly frame pointer. */
-	if (htt->target_version_major < 3 ||
-	    !ieee80211_is_mgmt(hdr->frame_control)) {
+	if (use_frags) {
 		/* tx fragment list must be terminated with zero-entry */
 		skb_put(txfrag, frag_len);
 		tx_frags = (struct htt_data_tx_desc_frag *)txfrag->data;
@@ -451,7 +450,7 @@ int ath10k_htt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 
 		res = ath10k_skb_map(dev, txfrag);
 		if (res)
-			goto err;
+			goto err_unmap_msdu;
 
 		ath10k_dbg(ATH10K_DBG_HTT, "txfrag 0x%llx\n",
 			   (unsigned long long) ATH10K_SKB_CB(txfrag)->paddr);
@@ -476,15 +475,11 @@ int ath10k_htt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 		flags0 |= HTT_DATA_TX_DESC_FLAGS0_NO_ENCRYPT;
 	flags0 |= HTT_DATA_TX_DESC_FLAGS0_MAC_HDR_PRESENT;
 
-	/* Since HTT 3.0 there is no separate mgmt tx command. However in case
-	 * of mgmt tx using TX_FRM there is not tx fragment list. Instead of tx
-	 * fragment list host driver specifies directly frame pointer. */
-	if (htt->target_version_major >= 3 &&
-	    ieee80211_is_mgmt(hdr->frame_control))
-		flags0 |= SM(ATH10K_HW_TXRX_MGMT,
+	if (use_frags)
+		flags0 |= SM(ATH10K_HW_TXRX_NATIVE_WIFI,
 			     HTT_DATA_TX_DESC_FLAGS0_PKT_TYPE);
 	else
-		flags0 |= SM(ATH10K_HW_TXRX_NATIVE_WIFI,
+		flags0 |= SM(ATH10K_HW_TXRX_MGMT,
 			     HTT_DATA_TX_DESC_FLAGS0_PKT_TYPE);
 
 	flags1  = 0;
@@ -493,14 +488,10 @@ int ath10k_htt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 	flags1 |= HTT_DATA_TX_DESC_FLAGS1_CKSUM_L3_OFFLOAD;
 	flags1 |= HTT_DATA_TX_DESC_FLAGS1_CKSUM_L4_OFFLOAD;
 
-	/* Since HTT 3.0 there is no separate mgmt tx command. However in case
-	 * of mgmt tx using TX_FRM there is not tx fragment list. Instead of tx
-	 * fragment list host driver specifies directly frame pointer. */
-	if (htt->target_version_major >= 3 &&
-	    ieee80211_is_mgmt(hdr->frame_control))
-		frags_paddr = ATH10K_SKB_CB(msdu)->paddr;
-	else
+	if (use_frags)
 		frags_paddr = ATH10K_SKB_CB(txfrag)->paddr;
+	else
+		frags_paddr = ATH10K_SKB_CB(msdu)->paddr;
 
 	cmd->hdr.msg_type        = HTT_H2T_MSG_TYPE_TX_FRM;
 	cmd->data_tx.flags0      = flags0;
@@ -514,23 +505,27 @@ int ath10k_htt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 
 	res = ath10k_htc_send(&htt->ar->htc, htt->eid, txdesc);
 	if (res)
-		goto err;
+		goto err_restore;
 
 	return 0;
-err:
-	if (txfrag)
+
+err_restore:
+	if (use_frags)
 		ath10k_skb_unmap(dev, txfrag);
-	if (txdesc)
-		dev_kfree_skb_any(txdesc);
-	if (txfrag)
+err_unmap_msdu:
+	ath10k_skb_unmap(dev, msdu);
+err_free_txfrag:
+	if (use_frags)
 		dev_kfree_skb_any(txfrag);
-	if (msdu_id >= 0) {
-		spin_lock_bh(&htt->tx_lock);
-		htt->pending_tx[msdu_id] = NULL;
-		ath10k_htt_tx_free_msdu_id(htt, msdu_id);
-		spin_unlock_bh(&htt->tx_lock);
-	}
+err_free_txdesc:
+	dev_kfree_skb_any(txdesc);
+err_free_msdu_id:
+	spin_lock_bh(&htt->tx_lock);
+	htt->pending_tx[msdu_id] = NULL;
+	ath10k_htt_tx_free_msdu_id(htt, msdu_id);
+	spin_unlock_bh(&htt->tx_lock);
+err_tx_dec:
 	ath10k_htt_tx_dec_pending(htt);
-	ath10k_skb_unmap(dev, msdu);
+err:
 	return res;
 }
-- 
1.7.9.5

[PATCH 5/5] ath10k: use msdu headroom to store txfrag

[Michal Kazior]

Instead of allocating sk_buff for a mere 16-byte
tx fragment list buffer use headroom of the
original msdu sk_buff.

This decreases CPU cache pressure and improves
performance.

Measured improvement on AP135 is 560mbps ->
590mbps of UDP TX briding traffic.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
---
 drivers/net/wireless/ath/ath10k/core.h   |    4 +-
 drivers/net/wireless/ath/ath10k/htt_tx.c |   82 +++++++++++++++---------------
 drivers/net/wireless/ath/ath10k/mac.c    |    4 ++
 drivers/net/wireless/ath/ath10k/txrx.c   |   16 +++---
 4 files changed, 53 insertions(+), 53 deletions(-)

diff --git a/drivers/net/wireless/ath/ath10k/core.h b/drivers/net/wireless/ath/ath10k/core.h
index 4563f80..292ad45 100644
--- a/drivers/net/wireless/ath/ath10k/core.h
+++ b/drivers/net/wireless/ath/ath10k/core.h
@@ -54,7 +54,9 @@ struct ath10k_skb_cb {
 		u8 vdev_id;
 		u8 tid;
 		bool is_offchan;
-		struct sk_buff *txfrag;
+
+		u8 frag_len;
+		u8 pad_len;
 	} __packed htt;
 } __packed;
 
diff --git a/drivers/net/wireless/ath/ath10k/htt_tx.c b/drivers/net/wireless/ath/ath10k/htt_tx.c
index 5e738d8..3b93c6a 100644
--- a/drivers/net/wireless/ath/ath10k/htt_tx.c
+++ b/drivers/net/wireless/ath/ath10k/htt_tx.c
@@ -307,7 +307,8 @@ int ath10k_htt_mgmt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 	struct device *dev = htt->ar->dev;
 	struct sk_buff *txdesc = NULL;
 	struct htt_cmd *cmd;
-	u8 vdev_id = ATH10K_SKB_CB(msdu)->htt.vdev_id;
+	struct ath10k_skb_cb *skb_cb = ATH10K_SKB_CB(msdu);
+	u8 vdev_id = skb_cb->htt.vdev_id;
 	int len = 0;
 	int msdu_id = -1;
 	int res;
@@ -350,6 +351,9 @@ int ath10k_htt_mgmt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 	memcpy(cmd->mgmt_tx.hdr, msdu->data,
 	       min_t(int, msdu->len, HTT_MGMT_FRM_HDR_DOWNLOAD_LEN));
 
+	skb_cb->htt.frag_len = 0;
+	skb_cb->htt.pad_len = 0;
+
 	res = ath10k_htc_send(&htt->ar->htc, htt->eid, txdesc);
 	if (res)
 		goto err_unmap_msdu;
@@ -377,13 +381,12 @@ int ath10k_htt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 	struct htt_cmd *cmd;
 	struct htt_data_tx_desc_frag *tx_frags;
 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)msdu->data;
+	struct ath10k_skb_cb *skb_cb = ATH10K_SKB_CB(msdu);
 	struct sk_buff *txdesc = NULL;
-	struct sk_buff *txfrag = NULL;
 	bool use_frags;
 	u8 vdev_id = ATH10K_SKB_CB(msdu)->htt.vdev_id;
 	u8 tid;
-	int prefetch_len, desc_len, frag_len;
-	dma_addr_t frags_paddr;
+	int prefetch_len, desc_len;
 	int msdu_id = -1;
 	int res;
 	u8 flags0;
@@ -407,7 +410,6 @@ int ath10k_htt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 	prefetch_len = roundup(prefetch_len, 4);
 
 	desc_len = sizeof(cmd->hdr) + sizeof(cmd->data_tx) + prefetch_len;
-	frag_len = sizeof(*tx_frags) * 2;
 
 	txdesc = ath10k_htc_alloc_skb(desc_len);
 	if (!txdesc) {
@@ -421,41 +423,44 @@ int ath10k_htt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 	use_frags = htt->target_version_major < 3 ||
 		    !ieee80211_is_mgmt(hdr->frame_control);
 
-	if (use_frags) {
-		txfrag = dev_alloc_skb(frag_len);
-		if (!txfrag) {
-			res = -ENOMEM;
-			goto err_free_txdesc;
-		}
-	}
-
 	if (!IS_ALIGNED((unsigned long)txdesc->data, 4)) {
 		ath10k_warn("htt alignment check failed. dropping packet.\n");
 		res = -EIO;
-		goto err_free_txfrag;
+		goto err_free_txdesc;
+	}
+
+	if (use_frags) {
+		skb_cb->htt.frag_len = sizeof(*tx_frags) * 2;
+		skb_cb->htt.pad_len = (unsigned long)msdu->data -
+				      round_down((unsigned long)msdu->data, 4);
+
+		skb_push(msdu, skb_cb->htt.frag_len + skb_cb->htt.pad_len);
+	} else {
+		skb_cb->htt.frag_len = 0;
+		skb_cb->htt.pad_len = 0;
 	}
 
 	res = ath10k_skb_map(dev, msdu);
 	if (res)
-		goto err_free_txfrag;
+		goto err_pull_txfrag;
 
 	if (use_frags) {
+		dma_sync_single_for_cpu(dev, skb_cb->paddr, msdu->len,
+					DMA_TO_DEVICE);
+
 		/* tx fragment list must be terminated with zero-entry */
-		skb_put(txfrag, frag_len);
-		tx_frags = (struct htt_data_tx_desc_frag *)txfrag->data;
-		tx_frags[0].paddr = __cpu_to_le32(ATH10K_SKB_CB(msdu)->paddr);
-		tx_frags[0].len   = __cpu_to_le32(msdu->len);
+		tx_frags = (struct htt_data_tx_desc_frag *)msdu->data;
+		tx_frags[0].paddr = __cpu_to_le32(skb_cb->paddr +
+						  skb_cb->htt.frag_len +
+						  skb_cb->htt.pad_len);
+		tx_frags[0].len   = __cpu_to_le32(msdu->len -
+						  skb_cb->htt.frag_len -
+						  skb_cb->htt.pad_len);
 		tx_frags[1].paddr = __cpu_to_le32(0);
 		tx_frags[1].len   = __cpu_to_le32(0);
 
-		res = ath10k_skb_map(dev, txfrag);
-		if (res)
-			goto err_unmap_msdu;
-
-		ath10k_dbg(ATH10K_DBG_HTT, "txfrag 0x%llx\n",
-			   (unsigned long long) ATH10K_SKB_CB(txfrag)->paddr);
-		ath10k_dbg_dump(ATH10K_DBG_HTT_DUMP, NULL, "txfrag: ",
-				txfrag->data, frag_len);
+		dma_sync_single_for_device(dev, skb_cb->paddr, msdu->len,
+					   DMA_TO_DEVICE);
 	}
 
 	ath10k_dbg(ATH10K_DBG_HTT, "msdu 0x%llx\n",
@@ -488,35 +493,28 @@ int ath10k_htt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 	flags1 |= HTT_DATA_TX_DESC_FLAGS1_CKSUM_L3_OFFLOAD;
 	flags1 |= HTT_DATA_TX_DESC_FLAGS1_CKSUM_L4_OFFLOAD;
 
-	if (use_frags)
-		frags_paddr = ATH10K_SKB_CB(txfrag)->paddr;
-	else
-		frags_paddr = ATH10K_SKB_CB(msdu)->paddr;
-
 	cmd->hdr.msg_type        = HTT_H2T_MSG_TYPE_TX_FRM;
 	cmd->data_tx.flags0      = flags0;
 	cmd->data_tx.flags1      = __cpu_to_le16(flags1);
-	cmd->data_tx.len         = __cpu_to_le16(msdu->len);
+	cmd->data_tx.len         = __cpu_to_le16(msdu->len -
+						 skb_cb->htt.frag_len -
+						 skb_cb->htt.pad_len);
 	cmd->data_tx.id          = __cpu_to_le16(msdu_id);
-	cmd->data_tx.frags_paddr = __cpu_to_le32(frags_paddr);
+	cmd->data_tx.frags_paddr = __cpu_to_le32(skb_cb->paddr);
 	cmd->data_tx.peerid      = __cpu_to_le32(HTT_INVALID_PEERID);
 
-	memcpy(cmd->data_tx.prefetch, msdu->data, prefetch_len);
+	memcpy(cmd->data_tx.prefetch, hdr, prefetch_len);
 
 	res = ath10k_htc_send(&htt->ar->htc, htt->eid, txdesc);
 	if (res)
-		goto err_restore;
+		goto err_unmap_msdu;
 
 	return 0;
 
-err_restore:
-	if (use_frags)
-		ath10k_skb_unmap(dev, txfrag);
 err_unmap_msdu:
 	ath10k_skb_unmap(dev, msdu);
-err_free_txfrag:
-	if (use_frags)
-		dev_kfree_skb_any(txfrag);
+err_pull_txfrag:
+	skb_pull(msdu, skb_cb->htt.frag_len + skb_cb->htt.pad_len);
 err_free_txdesc:
 	dev_kfree_skb_any(txdesc);
 err_free_msdu_id:
diff --git a/drivers/net/wireless/ath/ath10k/mac.c b/drivers/net/wireless/ath/ath10k/mac.c
index 9112e6d6f..99a9bad 100644
--- a/drivers/net/wireless/ath/ath10k/mac.c
+++ b/drivers/net/wireless/ath/ath10k/mac.c
@@ -3342,6 +3342,10 @@ int ath10k_mac_register(struct ath10k *ar)
 			IEEE80211_HW_WANT_MONITOR_VIF |
 			IEEE80211_HW_AP_LINK_PS;
 
+	/* MSDU can have HTT TX fragment pushed in front. The additional 4
+	 * bytes is used for padding/alignment if necessary. */
+	ar->hw->extra_tx_headroom += sizeof(struct htt_data_tx_desc_frag)*2 + 4;
+
 	if (ar->ht_cap_info & WMI_HT_CAP_DYNAMIC_SMPS)
 		ar->hw->flags |= IEEE80211_HW_SUPPORTS_DYNAMIC_SMPS;
 
diff --git a/drivers/net/wireless/ath/ath10k/txrx.c b/drivers/net/wireless/ath/ath10k/txrx.c
index 15395af..57931d0 100644
--- a/drivers/net/wireless/ath/ath10k/txrx.c
+++ b/drivers/net/wireless/ath/ath10k/txrx.c
@@ -49,7 +49,8 @@ void ath10k_txrx_tx_unref(struct ath10k_htt *htt,
 {
 	struct device *dev = htt->ar->dev;
 	struct ieee80211_tx_info *info;
-	struct sk_buff *msdu, *txfrag;
+	struct ath10k_skb_cb *skb_cb;
+	struct sk_buff *msdu;
 	int ret;
 
 	ath10k_dbg(ATH10K_DBG_HTT, "htt tx completion msdu_id %u discard %d no_ack %d\n",
@@ -62,20 +63,15 @@ void ath10k_txrx_tx_unref(struct ath10k_htt *htt,
 	}
 
 	msdu = htt->pending_tx[tx_done->msdu_id];
-	txfrag = ATH10K_SKB_CB(msdu)->htt.txfrag;
-
-	if (txfrag) {
-		ret = ath10k_skb_unmap(dev, txfrag);
-		if (ret)
-			ath10k_warn("txfrag unmap failed (%d)\n", ret);
-
-		dev_kfree_skb_any(txfrag);
-	}
+	skb_cb = ATH10K_SKB_CB(msdu);
 
 	ret = ath10k_skb_unmap(dev, msdu);
 	if (ret)
 		ath10k_warn("data skb unmap failed (%d)\n", ret);
 
+	if (skb_cb->htt.frag_len)
+		skb_pull(msdu, skb_cb->htt.frag_len + skb_cb->htt.pad_len);
+
 	ath10k_report_offchan_tx(htt->ar, msdu);
 
 	info = IEEE80211_SKB_CB(msdu);
-- 
1.7.9.5


_______________________________________________
ath10k mailing list
ath10k@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/ath10k

[PATCH 5/5] ath10k: use msdu headroom to store txfrag

[Michal Kazior]

Instead of allocating sk_buff for a mere 16-byte
tx fragment list buffer use headroom of the
original msdu sk_buff.

This decreases CPU cache pressure and improves
performance.

Measured improvement on AP135 is 560mbps ->
590mbps of UDP TX briding traffic.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
---
 drivers/net/wireless/ath/ath10k/core.h   |    4 +-
 drivers/net/wireless/ath/ath10k/htt_tx.c |   82 +++++++++++++++---------------
 drivers/net/wireless/ath/ath10k/mac.c    |    4 ++
 drivers/net/wireless/ath/ath10k/txrx.c   |   16 +++---
 4 files changed, 53 insertions(+), 53 deletions(-)

diff --git a/drivers/net/wireless/ath/ath10k/core.h b/drivers/net/wireless/ath/ath10k/core.h
index 4563f80..292ad45 100644
--- a/drivers/net/wireless/ath/ath10k/core.h
+++ b/drivers/net/wireless/ath/ath10k/core.h
@@ -54,7 +54,9 @@ struct ath10k_skb_cb {
 		u8 vdev_id;
 		u8 tid;
 		bool is_offchan;
-		struct sk_buff *txfrag;
+
+		u8 frag_len;
+		u8 pad_len;
 	} __packed htt;
 } __packed;
 
diff --git a/drivers/net/wireless/ath/ath10k/htt_tx.c b/drivers/net/wireless/ath/ath10k/htt_tx.c
index 5e738d8..3b93c6a 100644
--- a/drivers/net/wireless/ath/ath10k/htt_tx.c
+++ b/drivers/net/wireless/ath/ath10k/htt_tx.c
@@ -307,7 +307,8 @@ int ath10k_htt_mgmt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 	struct device *dev = htt->ar->dev;
 	struct sk_buff *txdesc = NULL;
 	struct htt_cmd *cmd;
-	u8 vdev_id = ATH10K_SKB_CB(msdu)->htt.vdev_id;
+	struct ath10k_skb_cb *skb_cb = ATH10K_SKB_CB(msdu);
+	u8 vdev_id = skb_cb->htt.vdev_id;
 	int len = 0;
 	int msdu_id = -1;
 	int res;
@@ -350,6 +351,9 @@ int ath10k_htt_mgmt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 	memcpy(cmd->mgmt_tx.hdr, msdu->data,
 	       min_t(int, msdu->len, HTT_MGMT_FRM_HDR_DOWNLOAD_LEN));
 
+	skb_cb->htt.frag_len = 0;
+	skb_cb->htt.pad_len = 0;
+
 	res = ath10k_htc_send(&htt->ar->htc, htt->eid, txdesc);
 	if (res)
 		goto err_unmap_msdu;
@@ -377,13 +381,12 @@ int ath10k_htt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 	struct htt_cmd *cmd;
 	struct htt_data_tx_desc_frag *tx_frags;
 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)msdu->data;
+	struct ath10k_skb_cb *skb_cb = ATH10K_SKB_CB(msdu);
 	struct sk_buff *txdesc = NULL;
-	struct sk_buff *txfrag = NULL;
 	bool use_frags;
 	u8 vdev_id = ATH10K_SKB_CB(msdu)->htt.vdev_id;
 	u8 tid;
-	int prefetch_len, desc_len, frag_len;
-	dma_addr_t frags_paddr;
+	int prefetch_len, desc_len;
 	int msdu_id = -1;
 	int res;
 	u8 flags0;
@@ -407,7 +410,6 @@ int ath10k_htt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 	prefetch_len = roundup(prefetch_len, 4);
 
 	desc_len = sizeof(cmd->hdr) + sizeof(cmd->data_tx) + prefetch_len;
-	frag_len = sizeof(*tx_frags) * 2;
 
 	txdesc = ath10k_htc_alloc_skb(desc_len);
 	if (!txdesc) {
@@ -421,41 +423,44 @@ int ath10k_htt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 	use_frags = htt->target_version_major < 3 ||
 		    !ieee80211_is_mgmt(hdr->frame_control);
 
-	if (use_frags) {
-		txfrag = dev_alloc_skb(frag_len);
-		if (!txfrag) {
-			res = -ENOMEM;
-			goto err_free_txdesc;
-		}
-	}
-
 	if (!IS_ALIGNED((unsigned long)txdesc->data, 4)) {
 		ath10k_warn("htt alignment check failed. dropping packet.\n");
 		res = -EIO;
-		goto err_free_txfrag;
+		goto err_free_txdesc;
+	}
+
+	if (use_frags) {
+		skb_cb->htt.frag_len = sizeof(*tx_frags) * 2;
+		skb_cb->htt.pad_len = (unsigned long)msdu->data -
+				      round_down((unsigned long)msdu->data, 4);
+
+		skb_push(msdu, skb_cb->htt.frag_len + skb_cb->htt.pad_len);
+	} else {
+		skb_cb->htt.frag_len = 0;
+		skb_cb->htt.pad_len = 0;
 	}
 
 	res = ath10k_skb_map(dev, msdu);
 	if (res)
-		goto err_free_txfrag;
+		goto err_pull_txfrag;
 
 	if (use_frags) {
+		dma_sync_single_for_cpu(dev, skb_cb->paddr, msdu->len,
+					DMA_TO_DEVICE);
+
 		/* tx fragment list must be terminated with zero-entry */
-		skb_put(txfrag, frag_len);
-		tx_frags = (struct htt_data_tx_desc_frag *)txfrag->data;
-		tx_frags[0].paddr = __cpu_to_le32(ATH10K_SKB_CB(msdu)->paddr);
-		tx_frags[0].len   = __cpu_to_le32(msdu->len);
+		tx_frags = (struct htt_data_tx_desc_frag *)msdu->data;
+		tx_frags[0].paddr = __cpu_to_le32(skb_cb->paddr +
+						  skb_cb->htt.frag_len +
+						  skb_cb->htt.pad_len);
+		tx_frags[0].len   = __cpu_to_le32(msdu->len -
+						  skb_cb->htt.frag_len -
+						  skb_cb->htt.pad_len);
 		tx_frags[1].paddr = __cpu_to_le32(0);
 		tx_frags[1].len   = __cpu_to_le32(0);
 
-		res = ath10k_skb_map(dev, txfrag);
-		if (res)
-			goto err_unmap_msdu;
-
-		ath10k_dbg(ATH10K_DBG_HTT, "txfrag 0x%llx\n",
-			   (unsigned long long) ATH10K_SKB_CB(txfrag)->paddr);
-		ath10k_dbg_dump(ATH10K_DBG_HTT_DUMP, NULL, "txfrag: ",
-				txfrag->data, frag_len);
+		dma_sync_single_for_device(dev, skb_cb->paddr, msdu->len,
+					   DMA_TO_DEVICE);
 	}
 
 	ath10k_dbg(ATH10K_DBG_HTT, "msdu 0x%llx\n",
@@ -488,35 +493,28 @@ int ath10k_htt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 	flags1 |= HTT_DATA_TX_DESC_FLAGS1_CKSUM_L3_OFFLOAD;
 	flags1 |= HTT_DATA_TX_DESC_FLAGS1_CKSUM_L4_OFFLOAD;
 
-	if (use_frags)
-		frags_paddr = ATH10K_SKB_CB(txfrag)->paddr;
-	else
-		frags_paddr = ATH10K_SKB_CB(msdu)->paddr;
-
 	cmd->hdr.msg_type        = HTT_H2T_MSG_TYPE_TX_FRM;
 	cmd->data_tx.flags0      = flags0;
 	cmd->data_tx.flags1      = __cpu_to_le16(flags1);
-	cmd->data_tx.len         = __cpu_to_le16(msdu->len);
+	cmd->data_tx.len         = __cpu_to_le16(msdu->len -
+						 skb_cb->htt.frag_len -
+						 skb_cb->htt.pad_len);
 	cmd->data_tx.id          = __cpu_to_le16(msdu_id);
-	cmd->data_tx.frags_paddr = __cpu_to_le32(frags_paddr);
+	cmd->data_tx.frags_paddr = __cpu_to_le32(skb_cb->paddr);
 	cmd->data_tx.peerid      = __cpu_to_le32(HTT_INVALID_PEERID);
 
-	memcpy(cmd->data_tx.prefetch, msdu->data, prefetch_len);
+	memcpy(cmd->data_tx.prefetch, hdr, prefetch_len);
 
 	res = ath10k_htc_send(&htt->ar->htc, htt->eid, txdesc);
 	if (res)
-		goto err_restore;
+		goto err_unmap_msdu;
 
 	return 0;
 
-err_restore:
-	if (use_frags)
-		ath10k_skb_unmap(dev, txfrag);
 err_unmap_msdu:
 	ath10k_skb_unmap(dev, msdu);
-err_free_txfrag:
-	if (use_frags)
-		dev_kfree_skb_any(txfrag);
+err_pull_txfrag:
+	skb_pull(msdu, skb_cb->htt.frag_len + skb_cb->htt.pad_len);
 err_free_txdesc:
 	dev_kfree_skb_any(txdesc);
 err_free_msdu_id:
diff --git a/drivers/net/wireless/ath/ath10k/mac.c b/drivers/net/wireless/ath/ath10k/mac.c
index 9112e6d6f..99a9bad 100644
--- a/drivers/net/wireless/ath/ath10k/mac.c
+++ b/drivers/net/wireless/ath/ath10k/mac.c
@@ -3342,6 +3342,10 @@ int ath10k_mac_register(struct ath10k *ar)
 			IEEE80211_HW_WANT_MONITOR_VIF |
 			IEEE80211_HW_AP_LINK_PS;
 
+	/* MSDU can have HTT TX fragment pushed in front. The additional 4
+	 * bytes is used for padding/alignment if necessary. */
+	ar->hw->extra_tx_headroom += sizeof(struct htt_data_tx_desc_frag)*2 + 4;
+
 	if (ar->ht_cap_info & WMI_HT_CAP_DYNAMIC_SMPS)
 		ar->hw->flags |= IEEE80211_HW_SUPPORTS_DYNAMIC_SMPS;
 
diff --git a/drivers/net/wireless/ath/ath10k/txrx.c b/drivers/net/wireless/ath/ath10k/txrx.c
index 15395af..57931d0 100644
--- a/drivers/net/wireless/ath/ath10k/txrx.c
+++ b/drivers/net/wireless/ath/ath10k/txrx.c
@@ -49,7 +49,8 @@ void ath10k_txrx_tx_unref(struct ath10k_htt *htt,
 {
 	struct device *dev = htt->ar->dev;
 	struct ieee80211_tx_info *info;
-	struct sk_buff *msdu, *txfrag;
+	struct ath10k_skb_cb *skb_cb;
+	struct sk_buff *msdu;
 	int ret;
 
 	ath10k_dbg(ATH10K_DBG_HTT, "htt tx completion msdu_id %u discard %d no_ack %d\n",
@@ -62,20 +63,15 @@ void ath10k_txrx_tx_unref(struct ath10k_htt *htt,
 	}
 
 	msdu = htt->pending_tx[tx_done->msdu_id];
-	txfrag = ATH10K_SKB_CB(msdu)->htt.txfrag;
-
-	if (txfrag) {
-		ret = ath10k_skb_unmap(dev, txfrag);
-		if (ret)
-			ath10k_warn("txfrag unmap failed (%d)\n", ret);
-
-		dev_kfree_skb_any(txfrag);
-	}
+	skb_cb = ATH10K_SKB_CB(msdu);
 
 	ret = ath10k_skb_unmap(dev, msdu);
 	if (ret)
 		ath10k_warn("data skb unmap failed (%d)\n", ret);
 
+	if (skb_cb->htt.frag_len)
+		skb_pull(msdu, skb_cb->htt.frag_len + skb_cb->htt.pad_len);
+
 	ath10k_report_offchan_tx(htt->ar, msdu);
 
 	info = IEEE80211_SKB_CB(msdu);
-- 
1.7.9.5

Re: [PATCH 0/5] ath10k: tx performance patches

[Kalle Valo]

Michal Kazior <michal.kazior@tieto.com> writes:

> Hi,
>
> This patchset reduces CPU load and improves
> improves TX performance on AP135 by additional
> 30mbps (give or take; 560mbps -> 590mbps), at
> least on my testbed.
>
>
> Michal Kazior (5):
>   ath10k: use num_pending_tx instead of msdu id bitmap
>   ath10k: avoid needless memset on TX path
>   ath10k: decouple HTT TX completions
>   ath10k: cleanup HTT TX functions
>   ath10k: use msdu headroom to store txfrag

Thanks, all five applied.

-- 
Kalle Valo

_______________________________________________
ath10k mailing list
ath10k@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/ath10k

Re: [PATCH 0/5] ath10k: tx performance patches

[Kalle Valo]

Michal Kazior <michal.kazior@tieto.com> writes:

> Hi,
>
> This patchset reduces CPU load and improves
> improves TX performance on AP135 by additional
> 30mbps (give or take; 560mbps -> 590mbps), at
> least on my testbed.
>
>
> Michal Kazior (5):
>   ath10k: use num_pending_tx instead of msdu id bitmap
>   ath10k: avoid needless memset on TX path
>   ath10k: decouple HTT TX completions
>   ath10k: cleanup HTT TX functions
>   ath10k: use msdu headroom to store txfrag

Thanks, all five applied.

-- 
Kalle Valo