[PATCH 0/2] ath10k: implement checksum offloading

[PATCH 0/2] ath10k: implement checksum offloading

[Michal Kazior]

Hi,

This patchset adds tx and rx checksum offloading
to ath10k driver.

I don't have any hard numbers but I did observe a
consistently sligthly better performance on AP135
+ ath10k. Nothing ground breaking though.


Michal Kazior (2):
  ath10k: implement rx checksum offloading
  ath10k: implement tx checksum offloading

 drivers/net/wireless/ath/ath10k/htt_rx.c |   40 ++++++++++++++++++++++++++++++
 drivers/net/wireless/ath/ath10k/htt_tx.c |    2 ++
 drivers/net/wireless/ath/ath10k/mac.c    |    2 ++
 3 files changed, 44 insertions(+)

-- 
1.7.9.5

[PATCH 1/2] ath10k: implement rx checksum offloading

[Michal Kazior]

HW supports L3/L4 rx checksum offloading.

This should reduce CPU load and improve
performance on slow host machines.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
---
 drivers/net/wireless/ath/ath10k/htt_rx.c |   40 ++++++++++++++++++++++++++++++
 1 file changed, 40 insertions(+)

diff --git a/drivers/net/wireless/ath/ath10k/htt_rx.c b/drivers/net/wireless/ath/ath10k/htt_rx.c
index 04f08d9..e784c40 100644
--- a/drivers/net/wireless/ath/ath10k/htt_rx.c
+++ b/drivers/net/wireless/ath/ath10k/htt_rx.c
@@ -804,6 +804,37 @@ static bool ath10k_htt_rx_has_fcs_err(struct sk_buff *skb)
 	return false;
 }
 
+static int ath10k_htt_rx_get_csum_state(struct sk_buff *skb)
+{
+	struct htt_rx_desc *rxd;
+	u32 flags, info;
+	bool is_ip4, is_ip6;
+	bool is_tcp, is_udp;
+	bool ip_csum_ok, tcpudp_csum_ok;
+
+	rxd = (void *)skb->data - sizeof(*rxd);
+	flags = __le32_to_cpu(rxd->attention.flags);
+	info = __le32_to_cpu(rxd->msdu_start.info1);
+
+	is_ip4 = !!(info & RX_MSDU_START_INFO1_IPV4_PROTO);
+	is_ip6 = !!(info & RX_MSDU_START_INFO1_IPV6_PROTO);
+	is_tcp = !!(info & RX_MSDU_START_INFO1_TCP_PROTO);
+	is_udp = !!(info & RX_MSDU_START_INFO1_UDP_PROTO);
+	ip_csum_ok = !(flags & RX_ATTENTION_FLAGS_IP_CHKSUM_FAIL);
+	tcpudp_csum_ok = !(flags & RX_ATTENTION_FLAGS_TCP_UDP_CHKSUM_FAIL);
+
+	if (!is_ip4 && !is_ip6)
+		return CHECKSUM_NONE;
+	if (!is_tcp && !is_udp)
+		return CHECKSUM_NONE;
+	if (!ip_csum_ok)
+		return CHECKSUM_NONE;
+	if (!tcpudp_csum_ok)
+		return CHECKSUM_NONE;
+
+	return CHECKSUM_UNNECESSARY;
+}
+
 static void ath10k_htt_rx_handler(struct ath10k_htt *htt,
 				  struct htt_rx_indication *rx)
 {
@@ -815,6 +846,7 @@ static void ath10k_htt_rx_handler(struct ath10k_htt *htt,
 	u8 *fw_desc;
 	int i, j;
 	int ret;
+	int ip_summed;
 
 	memset(&info, 0, sizeof(info));
 
@@ -889,6 +921,11 @@ static void ath10k_htt_rx_handler(struct ath10k_htt *htt,
 				continue;
 			}
 
+			/* The skb is not yet processed and it may be
+			 * reallocated. Since the offload is in the original
+			 * skb extract the checksum now and assign it later */
+			ip_summed = ath10k_htt_rx_get_csum_state(msdu_head);
+
 			info.skb     = msdu_head;
 			info.fcs_err = ath10k_htt_rx_has_fcs_err(msdu_head);
 			info.signal  = ATH10K_DEFAULT_NOISE_FLOOR;
@@ -914,6 +951,8 @@ static void ath10k_htt_rx_handler(struct ath10k_htt *htt,
 			if (ath10k_htt_rx_hdr_is_amsdu((void *)info.skb->data))
 				ath10k_dbg(ATH10K_DBG_HTT, "htt mpdu is amsdu\n");
 
+			info.skb->ip_summed = ip_summed;
+
 			ath10k_dbg_dump(ATH10K_DBG_HTT_DUMP, NULL, "htt mpdu: ",
 					info.skb->data, info.skb->len);
 			ath10k_process_rx(htt->ar, &info);
@@ -980,6 +1019,7 @@ static void ath10k_htt_rx_frag_handler(struct ath10k_htt *htt,
 	info.status = HTT_RX_IND_MPDU_STATUS_OK;
 	info.encrypt_type = MS(__le32_to_cpu(rxd->mpdu_start.info0),
 				RX_MPDU_START_INFO0_ENCRYPT_TYPE);
+	info.skb->ip_summed = ath10k_htt_rx_get_csum_state(info.skb);
 
 	if (tkip_mic_err) {
 		ath10k_warn("tkip mic error\n");
-- 
1.7.9.5

[PATCH 2/2] ath10k: implement tx checksum offloading

[Michal Kazior]

HW supports L3/L4 tx checksum offloading.

This should reduce CPU load and improve
performance on slow host machines.

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

diff --git a/drivers/net/wireless/ath/ath10k/htt_tx.c b/drivers/net/wireless/ath/ath10k/htt_tx.c
index dc3f3e8..656c254 100644
--- a/drivers/net/wireless/ath/ath10k/htt_tx.c
+++ b/drivers/net/wireless/ath/ath10k/htt_tx.c
@@ -465,6 +465,8 @@ int ath10k_htt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
 	flags1  = 0;
 	flags1 |= SM((u16)vdev_id, HTT_DATA_TX_DESC_FLAGS1_VDEV_ID);
 	flags1 |= SM((u16)tid, HTT_DATA_TX_DESC_FLAGS1_EXT_TID);
+	flags1 |= HTT_DATA_TX_DESC_FLAGS1_CKSUM_L3_OFFLOAD;
+	flags1 |= HTT_DATA_TX_DESC_FLAGS1_CKSUM_L4_OFFLOAD;
 
 	frags_paddr = ATH10K_SKB_CB(txfrag)->paddr;
 
diff --git a/drivers/net/wireless/ath/ath10k/mac.c b/drivers/net/wireless/ath/ath10k/mac.c
index 344ad27..5784337 100644
--- a/drivers/net/wireless/ath/ath10k/mac.c
+++ b/drivers/net/wireless/ath/ath10k/mac.c
@@ -3309,6 +3309,8 @@ int ath10k_mac_register(struct ath10k *ar)
 	ar->hw->wiphy->iface_combinations = &ath10k_if_comb;
 	ar->hw->wiphy->n_iface_combinations = 1;
 
+	ar->hw->netdev_features = NETIF_F_HW_CSUM;
+
 	ret = ath_regd_init(&ar->ath_common.regulatory, ar->hw->wiphy,
 			    ath10k_reg_notifier);
 	if (ret) {
-- 
1.7.9.5

Re: [PATCH 0/2] ath10k: implement checksum offloading

[Kalle Valo]

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

> Hi,
>
> This patchset adds tx and rx checksum offloading
> to ath10k driver.
>
> I don't have any hard numbers but I did observe a
> consistently sligthly better performance on AP135
> + ath10k. Nothing ground breaking though.
>
>
> Michal Kazior (2):
>   ath10k: implement rx checksum offloading
>   ath10k: implement tx checksum offloading

Applied, thanks.

-- 
Kalle Valo