ar71xx: add some hacks to work around the misalignment in IP packets received on...

decreases CPU load with the default firewall for routing 95 mbit/s from 78% to 55%

git-svn-id: svn://svn.openwrt.org/openwrt/trunk@27878 3c298f89-4303-0410-b956-a3cf2f4a3e73

diff --git a/target/linux/ar71xx/patches-2.6.39/910-unaligned_access_hacks.patch b/target/linux/ar71xx/patches-2.6.39/910-unaligned_access_hacks.patch
new file mode 100644 (file)
index 0000000..921cf19
--- /dev/null
+++ b/target/linux/ar71xx/patches-2.6.39/910-unaligned_access_hacks.patch
@@ -0,0 +1,117 @@
+--- a/arch/mips/include/asm/checksum.h
++++ b/arch/mips/include/asm/checksum.h
+@@ -12,6 +12,7 @@
+ #define _ASM_CHECKSUM_H
+ 
+ #include <linux/in6.h>
++#include <linux/unaligned/packed_struct.h>
+ 
+ #include <asm/uaccess.h>
+ 
+@@ -104,26 +105,30 @@ static inline __sum16 ip_fast_csum(const
+       const unsigned int *stop = word + ihl;
+       unsigned int csum;
+       int carry;
++      unsigned int w;
+ 
+-      csum = word[0];
+-      csum += word[1];
+-      carry = (csum < word[1]);
++      csum = __get_unaligned_cpu32(word++);
++
++      w = __get_unaligned_cpu32(word++);
++      csum += w;
++      carry = (csum < w);
+       csum += carry;
+ 
+-      csum += word[2];
+-      carry = (csum < word[2]);
++      w = __get_unaligned_cpu32(word++);
++      csum += w;
++      carry = (csum < w);
+       csum += carry;
+ 
+-      csum += word[3];
+-      carry = (csum < word[3]);
++      w = __get_unaligned_cpu32(word++);
++      csum += w;
++      carry = (csum < w);
+       csum += carry;
+ 
+-      word += 4;
+       do {
+-              csum += *word;
+-              carry = (csum < *word);
++              w = __get_unaligned_cpu32(word++);
++              csum += w;
++              carry = (csum < w);
+               csum += carry;
+-              word++;
+       } while (word != stop);
+ 
+       return csum_fold(csum);
+--- a/include/linux/ip.h
++++ b/include/linux/ip.h
+@@ -102,7 +102,7 @@ struct iphdr {
+       __be32  saddr;
+       __be32  daddr;
+       /*The options start here. */
+-};
++} __packed;
+ 
+ #ifdef __KERNEL__
+ #include <linux/skbuff.h>
+--- a/include/linux/ipv6.h
++++ b/include/linux/ipv6.h
+@@ -126,7 +126,7 @@ struct ipv6hdr {
+ 
+       struct  in6_addr        saddr;
+       struct  in6_addr        daddr;
+-};
++} __packed;
+ 
+ #ifdef __KERNEL__
+ /*
+--- a/include/linux/tcp.h
++++ b/include/linux/tcp.h
+@@ -54,7 +54,7 @@ struct tcphdr {
+       __be16  window;
+       __sum16 check;
+       __be16  urg_ptr;
+-};
++} __packed;
+ 
+ /*
+  *    The union cast uses a gcc extension to avoid aliasing problems
+--- a/include/linux/udp.h
++++ b/include/linux/udp.h
+@@ -24,7 +24,7 @@ struct udphdr {
+       __be16  dest;
+       __be16  len;
+       __sum16 check;
+-};
++} __packed;
+ 
+ /* UDP socket options */
+ #define UDP_CORK      1       /* Never send partially complete segments */
+--- a/net/ipv4/netfilter/nf_conntrack_l3proto_ipv4.c
++++ b/net/ipv4/netfilter/nf_conntrack_l3proto_ipv4.c
+@@ -14,6 +14,7 @@
+ #include <linux/skbuff.h>
+ #include <linux/icmp.h>
+ #include <linux/sysctl.h>
++#include <linux/unaligned/packed_struct.h>
+ #include <net/route.h>
+ #include <net/ip.h>
+ 
+@@ -44,8 +45,8 @@ static bool ipv4_pkt_to_tuple(const stru
+       if (ap == NULL)
+               return false;
+ 
+-      tuple->src.u3.ip = ap[0];
+-      tuple->dst.u3.ip = ap[1];
++      tuple->src.u3.ip = __get_unaligned_cpu32(ap++);
++      tuple->dst.u3.ip = __get_unaligned_cpu32(ap);
+ 
+       return true;
+ }