Avoid unaligned memory accesses in VXLAN and batman-adv

Improves performance slightly.
This commit is contained in:
Matthias Schiffer 2018-01-24 22:13:30 +01:00
parent 2950cc3f59
commit a9edd43693
No known key found for this signature in database
GPG Key ID: 16EF3F64CB201D9C
2 changed files with 177 additions and 0 deletions

View File

@ -0,0 +1,130 @@
From: Matthias Schiffer <mschiffer@universe-factory.net>
Date: Wed, 24 Jan 2018 21:34:54 +0100
Subject: ar71xx: add unaligned access hacks for VXLAN
Signed-off-by: Matthias Schiffer <mschiffer@universe-factory.net>
diff --git a/target/linux/ar71xx/patches-4.4/910-unaligned_access_hacks.patch b/target/linux/ar71xx/patches-4.4/910-unaligned_access_hacks.patch
index 8882b1758fed7ac254116fb0cb0e1b95bbe5259f..b06b36317f6bae7827c280ffc7338b6487d762f5 100644
--- a/target/linux/ar71xx/patches-4.4/910-unaligned_access_hacks.patch
+++ b/target/linux/ar71xx/patches-4.4/910-unaligned_access_hacks.patch
@@ -929,3 +929,119 @@
};
if (skb->ip_summed != CHECKSUM_PARTIAL) {
*sum = csum_fold(csum_partial(diff, sizeof(diff),
+--- a/drivers/net/vxlan.c
++++ b/drivers/net/vxlan.c
+@@ -1759,8 +1759,8 @@ static int vxlan6_xmit_skb(struct dst_en
+ }
+
+ vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
+- vxh->vx_flags = htonl(VXLAN_HF_VNI);
+- vxh->vx_vni = vni;
++ net_hdr_word(&vxh->vx_flags) = htonl(VXLAN_HF_VNI);
++ net_hdr_word(&vxh->vx_vni) = vni;
+
+ if (type & SKB_GSO_TUNNEL_REMCSUM) {
+ u32 data = (skb_checksum_start_offset(skb) - hdrlen) >>
+@@ -1769,8 +1769,8 @@ static int vxlan6_xmit_skb(struct dst_en
+ if (skb->csum_offset == offsetof(struct udphdr, check))
+ data |= VXLAN_RCO_UDP;
+
+- vxh->vx_vni |= htonl(data);
+- vxh->vx_flags |= htonl(VXLAN_HF_RCO);
++ net_hdr_word(&vxh->vx_vni) |= htonl(data);
++ net_hdr_word(&vxh->vx_flags) |= htonl(VXLAN_HF_RCO);
+
+ if (!skb_is_gso(skb)) {
+ skb->ip_summed = CHECKSUM_NONE;
+@@ -1838,8 +1838,8 @@ static int vxlan_xmit_skb(struct rtable
+ return PTR_ERR(skb);
+
+ vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
+- vxh->vx_flags = htonl(VXLAN_HF_VNI);
+- vxh->vx_vni = vni;
++ net_hdr_word(&vxh->vx_flags) = htonl(VXLAN_HF_VNI);
++ net_hdr_word(&vxh->vx_vni) = vni;
+
+ if (type & SKB_GSO_TUNNEL_REMCSUM) {
+ u32 data = (skb_checksum_start_offset(skb) - hdrlen) >>
+@@ -1848,8 +1848,8 @@ static int vxlan_xmit_skb(struct rtable
+ if (skb->csum_offset == offsetof(struct udphdr, check))
+ data |= VXLAN_RCO_UDP;
+
+- vxh->vx_vni |= htonl(data);
+- vxh->vx_flags |= htonl(VXLAN_HF_RCO);
++ net_hdr_word(&vxh->vx_vni) |= htonl(data);
++ net_hdr_word(&vxh->vx_flags) |= htonl(VXLAN_HF_RCO);
+
+ if (!skb_is_gso(skb)) {
+ skb->ip_summed = CHECKSUM_NONE;
+--- a/include/linux/etherdevice.h
++++ b/include/linux/etherdevice.h
+@@ -409,7 +409,7 @@ static inline bool is_etherdev_addr(cons
+ * @b: Pointer to Ethernet header
+ *
+ * Compare two Ethernet headers, returns 0 if equal.
+- * This assumes that the network header (i.e., IP header) is 4-byte
++ * This assumes that the network header (i.e., IP header) is 2-byte
+ * aligned OR the platform can handle unaligned access. This is the
+ * case for all packets coming into netif_receive_skb or similar
+ * entry points.
+@@ -432,11 +432,12 @@ static inline unsigned long compare_ethe
+ fold |= *(unsigned long *)(a + 6) ^ *(unsigned long *)(b + 6);
+ return fold;
+ #else
+- u32 *a32 = (u32 *)((u8 *)a + 2);
+- u32 *b32 = (u32 *)((u8 *)b + 2);
++ const u16 *a16 = a;
++ const u16 *b16 = b;
+
+- return (*(u16 *)a ^ *(u16 *)b) | (a32[0] ^ b32[0]) |
+- (a32[1] ^ b32[1]) | (a32[2] ^ b32[2]);
++ return (a16[0] ^ b16[0]) | (a16[1] ^ b16[1]) | (a16[2] ^ b16[2]) |
++ (a16[3] ^ b16[3]) | (a16[4] ^ b16[4]) | (a16[5] ^ b16[5]) |
++ (a16[6] ^ b16[6]);
+ #endif
+ }
+
+--- a/net/ipv4/tcp_offload.c
++++ b/net/ipv4/tcp_offload.c
+@@ -221,7 +221,7 @@ struct sk_buff **tcp_gro_receive(struct
+
+ th2 = tcp_hdr(p);
+
+- if (*(u32 *)&th->source ^ *(u32 *)&th2->source) {
++ if (net_hdr_word(&th->source) ^ net_hdr_word(&th2->source)) {
+ NAPI_GRO_CB(p)->same_flow = 0;
+ continue;
+ }
+@@ -239,8 +239,8 @@ found:
+ ~(TCP_FLAG_CWR | TCP_FLAG_FIN | TCP_FLAG_PSH));
+ flush |= (__force int)(th->ack_seq ^ th2->ack_seq);
+ for (i = sizeof(*th); i < thlen; i += 4)
+- flush |= *(u32 *)((u8 *)th + i) ^
+- *(u32 *)((u8 *)th2 + i);
++ flush |= net_hdr_word((u8 *)th + i) ^
++ net_hdr_word((u8 *)th2 + i);
+
+ mss = skb_shinfo(p)->gso_size;
+
+--- a/net/ipv6/netfilter/ip6table_mangle.c
++++ b/net/ipv6/netfilter/ip6table_mangle.c
+@@ -55,7 +55,7 @@ ip6t_mangle_out(struct sk_buff *skb, con
+ hop_limit = ipv6_hdr(skb)->hop_limit;
+
+ /* flowlabel and prio (includes version, which shouldn't change either */
+- flowlabel = *((u_int32_t *)ipv6_hdr(skb));
++ flowlabel = net_hdr_word(ipv6_hdr(skb));
+
+ ret = ip6t_do_table(skb, state, state->net->ipv6.ip6table_mangle);
+
+@@ -64,7 +64,7 @@ ip6t_mangle_out(struct sk_buff *skb, con
+ !ipv6_addr_equal(&ipv6_hdr(skb)->daddr, &daddr) ||
+ skb->mark != mark ||
+ ipv6_hdr(skb)->hop_limit != hop_limit ||
+- flowlabel != *((u_int32_t *)ipv6_hdr(skb)))) {
++ flowlabel != net_hdr_word(ipv6_hdr(skb)))) {
+ err = ip6_route_me_harder(state->net, skb);
+ if (err < 0)
+ ret = NF_DROP_ERR(err);

View File

@ -0,0 +1,47 @@
From: Matthias Schiffer <mschiffer@universe-factory.net>
Date: Tue, 23 Jan 2018 21:18:34 +0100
Subject: batman-adv: unaligned access fixes
Signed-off-by: Matthias Schiffer <mschiffer@universe-factory.net>
diff --git a/batman-adv/patches/900-unaligned-access.patch b/batman-adv/patches/900-unaligned-access.patch
new file mode 100644
index 0000000000000000000000000000000000000000..ba428b683ab146773162b0956a7e3b6260c14339
--- /dev/null
+++ b/batman-adv/patches/900-unaligned-access.patch
@@ -0,0 +1,35 @@
+--- a/net/batman-adv/packet.h
++++ b/net/batman-adv/packet.h
+@@ -173,8 +173,6 @@ struct batadv_bla_claim_dst {
+ __be16 group; /* group id */
+ };
+
+-#pragma pack()
+-
+ /**
+ * struct batadv_ogm_packet - ogm (routing protocol) packet
+ * @packet_type: batman-adv packet type, part of the general header
+@@ -382,7 +380,6 @@ struct batadv_icmp_packet_rr {
+ * misalignment of the payload after the ethernet header. It may also lead to
+ * leakage of information when the padding it not initialized before sending.
+ */
+-#pragma pack(2)
+
+ /**
+ * struct batadv_unicast_packet - unicast packet for network payload
+@@ -510,8 +507,6 @@ struct batadv_coded_packet {
+ __be16 coded_len;
+ };
+
+-#pragma pack()
+-
+ /**
+ * struct batadv_unicast_tvlv_packet - generic unicast packet with tvlv payload
+ * @packet_type: batman-adv packet type, part of the general header
+@@ -618,4 +613,6 @@ struct batadv_tvlv_mcast_data {
+ u8 reserved[3];
+ };
+
++#pragma pack()
++
+ #endif /* _NET_BATMAN_ADV_PACKET_H_ */