commit 4af1a7f938207f713c249270ed98a0c12f72cd3f
author Benoît Ganne <bganne@cisco.com> Fri Mar 01 14:14:10 2019 +0100
committer Damjan Marion <dmarion@me.com> Tue Mar 26 10:06:57 2019 +0000
tree 42c93b22e20228641de844d553819f400676cc5e
parent 50d5069a80fd763f4270d61d991febbdfa88aed3

Simplify adjacency rewrite code

Using memcpy instead of complex specific copy logic. This simplify
the implementation and also improve perf slightly.
Also move adjacency data from tail to head of buffer, which improves
cache locality (header and data share the same cacheline)
Finally, fix VxLAN which used to workaround vnet_rewrite logic.

Change-Id: I770ddad9846f7ee505aa99ad417e6a61d5cbbefa
Signed-off-by: Benoît Ganne <bganne@cisco.com>

src/vnet/vxlan/encap.c

diff --git a/src/vnet/vxlan/encap.c b/src/vnet/vxlan/encap.c
index fe2374c..b797624 100644
--- a/src/vnet/vxlan/encap.c
+++ b/src/vnet/vxlan/encap.c
@@ -92,7 +92,6 @@
 
   u8 const underlay_hdr_len = is_ip4 ?
     sizeof(ip4_vxlan_header_t) : sizeof(ip6_vxlan_header_t);
-  u8 const rw_hdr_offset = sizeof t0->rewrite_data - underlay_hdr_len;
   u16 const l3_len = is_ip4 ? sizeof(ip4_header_t) : sizeof(ip6_header_t);
   u32 const csum_flags = is_ip4 ?
     VNET_BUFFER_F_OFFLOAD_IP_CKSUM | VNET_BUFFER_F_IS_IP4 |
@@ -173,6 +172,7 @@
 
           ASSERT(t0->rewrite_header.data_bytes == underlay_hdr_len);
           ASSERT(t1->rewrite_header.data_bytes == underlay_hdr_len);
+          vnet_rewrite_two_headers(*t0, *t1, vlib_buffer_get_current(b0), vlib_buffer_get_current(b1), underlay_hdr_len);
 
           vlib_buffer_advance (b0, -underlay_hdr_len);
           vlib_buffer_advance (b1, -underlay_hdr_len);
@@ -185,12 +185,6 @@
           void * underlay0 = vlib_buffer_get_current(b0);
           void * underlay1 = vlib_buffer_get_current(b1);
 
-	  /* vnet_rewrite_two_header writes only in (uword) 8 bytes chunks
-           * and discards the first 4 bytes of the (36 bytes ip4 underlay)  rewrite
-           * use memcpy as a workaround */
-          clib_memcpy_fast(underlay0, t0->rewrite_header.data + rw_hdr_offset, underlay_hdr_len);
-          clib_memcpy_fast(underlay1, t1->rewrite_header.data + rw_hdr_offset, underlay_hdr_len);
-
           ip4_header_t * ip4_0, * ip4_1;
 	  qos_bits_t ip4_0_tos = 0, ip4_1_tos = 0;
           ip6_header_t * ip6_0, * ip6_1;
@@ -354,15 +348,11 @@
 	  vnet_buffer(b0)->ip.adj_index[VLIB_TX] = dpoi_idx0;
 
           ASSERT(t0->rewrite_header.data_bytes == underlay_hdr_len);
+          vnet_rewrite_one_header(*t0, vlib_buffer_get_current(b0), underlay_hdr_len);
 
           vlib_buffer_advance (b0, -underlay_hdr_len);
           void * underlay0 = vlib_buffer_get_current(b0);
 
-	  /* vnet_rewrite_one_header writes only in (uword) 8 bytes chunks
-           * and discards the first 4 bytes of the (36 bytes ip4 underlay)  rewrite
-           * use memcpy as a workaround */
-          clib_memcpy_fast(underlay0, t0->rewrite_header.data + rw_hdr_offset, underlay_hdr_len);
-
  	  u32 len0 = vlib_buffer_length_in_chain (vm, b0);
           u16 payload_l0 = clib_host_to_net_u16 (len0 - l3_len);