Skip to contentSkip to conversation

r8169: fix offloaded tx checksum for small packets.

PATCH In review Francois Romieu 3 participants 5 messages May 18, 2013 – Jun 10, 2013

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 r8169: fix offloaded tx checksum for small packets.

8168evl offloaded checksums are wrong since commit e5195c1f31f399289347e043d6abf3ffa80f0005 ("r8169: fix 8168evl frame padding.") pads small packets to 60 bytes (without ethernet checksum). Typical symptoms appear as UDP checksums which are wrong by the count of added bytes.

It isn't worth compensating. Let the driver checksum.

Due to the skb length changes, TSO code is moved before the Tx descriptor gets written.

---

I reworked some code since v1 to avoid wtf effects in the future. Holger's Tested-by is thus not included. While I trust my testing, it would be nice to wait for his ack as well.

This is obviously stuff for -stable.

 drivers/net/ethernet/realtek/r8169.c | 41 ++++++++++++++++++++++++------------
 1 file changed, 27 insertions(+), 14 deletions(-)
drivers/net/ethernet/realtek/r8169.c +27 -14
index 79c520b..393f961 100644
@@ -5856,7 +5856,20 @@ err_out:
 	return -EIO;
 }
 
-static inline void rtl8169_tso_csum(struct rtl8169_private *tp,
+static bool rtl_skb_pad(struct sk_buff *skb)
+{
+	if (skb_padto(skb, ETH_ZLEN))
+		return false;
+	skb_put(skb, ETH_ZLEN - skb->len);
+	return true;
+}
+
+static bool rtl_test_hw_pad_bug(struct rtl8169_private *tp, struct sk_buff *skb)
+{
+	return skb->len < ETH_ZLEN && tp->mac_version == RTL_GIGA_MAC_VER_34;
+}
+
+static inline bool rtl8169_tso_csum(struct rtl8169_private *tp,
 				    struct sk_buff *skb, u32 *opts)
 {
 	const struct rtl_tx_desc_info *info = tx_desc_info + tp->txd_version;
@@ -5869,13 +5882,20 @@ static inline void rtl8169_tso_csum(struct rtl8169_private *tp,
 	} else if (skb->ip_summed == CHECKSUM_PARTIAL) {
 		const struct iphdr *ip = ip_hdr(skb);
 
+		if (unlikely(rtl_test_hw_pad_bug(tp, skb)))
+			return skb_checksum_help(skb) == 0 && rtl_skb_pad(skb);
+
 		if (ip->protocol == IPPROTO_TCP)
 			opts[offset] |= info->checksum.tcp;
 		else if (ip->protocol == IPPROTO_UDP)
 			opts[offset] |= info->checksum.udp;
 		else
 			WARN_ON_ONCE(1);
+	} else {
+		if (unlikely(rtl_test_hw_pad_bug(tp, skb)))
+			return rtl_skb_pad(skb);
 	}
+	return true;
 }
 
 static netdev_tx_t rtl8169_start_xmit(struct sk_buff *skb,
@@ -5896,17 +5916,15 @@ static netdev_tx_t rtl8169_start_xmit(struct sk_buff *skb,
 		goto err_stop_0;
 	}
 
-	/* 8168evl does not automatically pad to minimum length. */
-	if (unlikely(tp->mac_version == RTL_GIGA_MAC_VER_34 &&
-		     skb->len < ETH_ZLEN)) {
-		if (skb_padto(skb, ETH_ZLEN))
-			goto err_update_stats;
-		skb_put(skb, ETH_ZLEN - skb->len);
-	}
-
 	if (unlikely(le32_to_cpu(txd->opts1) & DescOwn))
 		goto err_stop_0;
 
+	opts[1] = cpu_to_le32(rtl8169_tx_vlan_tag(skb));
+	opts[0] = DescOwn;
+
+	if (!rtl8169_tso_csum(tp, skb, opts))
+		goto err_update_stats;
+
 	len = skb_headlen(skb);
 	mapping = dma_map_single(d, skb->data, len, DMA_TO_DEVICE);
 	if (unlikely(dma_mapping_error(d, mapping))) {
@@ -5918,11 +5936,6 @@ static netdev_tx_t rtl8169_start_xmit(struct sk_buff *skb,
 	tp->tx_skb[entry].len = len;
 	txd->addr = cpu_to_le64(mapping);
 
-	opts[1] = cpu_to_le32(rtl8169_tx_vlan_tag(skb));
-	opts[0] = DescOwn;
-
-	rtl8169_tso_csum(tp, skb, opts);
-
 	frags = rtl8169_xmit_frags(tp, skb, opts);
 	if (frags < 0)
 		goto err_dma_1;
Signature · 1 line
1.8.1.4
Tested-by on v2 1/1
#1 lore raw
Quoted 11 lines · Francois Romieu

8168evl offloaded checksums are wrong since commit e5195c1f31f399289347e043d6abf3ffa80f0005 ("r8169: fix 8168evl frame padding.") pads small packets to 60 bytes (without ethernet checksum). Typical symptoms appear as UDP checksums which are wrong by the count of added bytes.

It isn't worth compensating. Let the driver checksum.

Due to the skb length changes, TSO code is moved before the Tx descriptor gets written.

Tested again, still working for me.

-h

#2 lore raw
Quoted 8 lines · Holger Hoffstätte
Quoted 7 lines · Francois Romieu

8168evl offloaded checksums are wrong since commit e5195c1f31f399289347e043d6abf3ffa80f0005 ("r8169: fix 8168evl frame padding.") pads small packets to 60 bytes (without ethernet checksum). Typical symptoms appear as UDP checksums which are wrong by the count of added bytes.

It isn't worth compensating. Let the driver checksum.

Quoted 8 lines
Quoted 4 lines

Due to the skb length changes, TSO code is moved before the Tx descriptor gets written.

Tested again, still working for me.

ping:

As far as I can tell this patch is still not in -stable 3.9.5 or queued up for the next iteration.

Commit in Linus' tree is b423e9ae49d78ea3f53b131c8d5a6087aed16fd6

thanks :)

-h

maintainer
#4 lore raw
In reply to Francois Romieu

From: Francois Romieu <[email protected]>
Date: Sat, 18 May 2013 13:24:46 +0200

Quoted 10 lines

8168evl offloaded checksums are wrong since commit e5195c1f31f399289347e043d6abf3ffa80f0005 ("r8169: fix 8168evl frame padding.") pads small packets to 60 bytes (without ethernet checksum). Typical symptoms appear as UDP checksums which are wrong by the count of added bytes.

It isn't worth compensating. Let the driver checksum.

Due to the skb length changes, TSO code is moved before the Tx descriptor gets written.

Quoted 8 lines

---

I reworked some code since v1 to avoid wtf effects in the future. Holger's Tested-by is thus not included. While I trust my testing, it would be nice to wait for his ack as well.

This is obviously stuff for -stable.

Applied and queued up for -stable, thanks!