VPP-106: fix performance hit due to unprefetched data read

Change-Id: I1325b60b6deadcb51631e178011a31ee70c06cc7
Signed-off-by: Dave Barach <dave@barachs.net>

4 files changed, 17 insertions, 47 deletions

diff --git a/dpdk/dpdk-16.04_patches/0017-i40e-Set-PKT_RX_VLAN_PKT-flag-for-VLAN-packets.patch b/dpdk/dpdk-16.04_patches/0017-i40e-Set-PKT_RX_VLAN_PKT-flag-for-VLAN-packets.patch
deleted file mode 100755
index f8f69b0f274..00000000000
--- a/dpdk/dpdk-16.04_patches/0017-i40e-Set-PKT_RX_VLAN_PKT-flag-for-VLAN-packets.patch
+++ /dev/null
@@ -1,41 +0,0 @@
-From af8bfd030fdaa0acb3222296ab7a24bc4ac196f5 Mon Sep 17 00:00:00 2001
-From: Georgi Savov <gsavov@cisco.com>
-Date: Tue, 10 May 2016 10:59:55 -0400
-Subject: [PATCH 30/31] i40e Set PKT_RX_VLAN_PKT flag for VLAN packets
-
----
- drivers/net/i40e/i40e_rxtx.c | 18 +++++++++++++++++-
- 1 file changed, 17 insertions(+), 1 deletion(-)
-
-diff --git a/drivers/net/i40e/i40e_rxtx.c b/drivers/net/i40e/i40e_rxtx.c
-index 39d94ec..07341df 100644
---- a/drivers/net/i40e/i40e_rxtx.c
-+++ b/drivers/net/i40e/i40e_rxtx.c
-@@ -99,7 +99,23 @@ i40e_rxd_to_vlan_tci(struct rte_mbuf *mb, volatile union i40e_rx_desc *rxdp)
- 		PMD_RX_LOG(DEBUG, "Descriptor l2tag1: %u",
- 			   rte_le_to_cpu_16(rxdp->wb.qword0.lo_dword.l2tag1));
- 	} else {
--		mb->vlan_tci = 0;
-+		struct ether_hdr *eth_hdr = rte_pktmbuf_mtod(mb, struct ether_hdr *);
-+		u16 eth_type = eth_hdr->ether_type;
-+
-+		// The i40e firmware does not flag VLAN tagged packets if
-+		// VLAN stripping is disabled so we need to check the
-+		// ethernet header to find out if the received packet
-+		// is a VLAN packet
-+		if ((eth_type == rte_be_to_cpu_16(ETHER_TYPE_VLAN))      ||
-+		    (eth_type == rte_be_to_cpu_16(ETHER_TYPE_VLAN_AD))   ||
-+		    (eth_type == rte_be_to_cpu_16(ETHER_TYPE_VLAN_9100)) ||
-+		    (eth_type == rte_be_to_cpu_16(ETHER_TYPE_VLAN_9200))) {
-+			struct vlan_hdr *vhdr = (struct vlan_hdr *)(eth_hdr+1);
-+			mb->ol_flags |= PKT_RX_VLAN_PKT;
-+			mb->vlan_tci = rte_be_to_cpu_16(vhdr->vlan_tci);
-+		} else {
-+			mb->vlan_tci = 0;
-+		}
- 	}
- #ifndef RTE_LIBRTE_I40E_16BYTE_RX_DESC
- 	if (rte_le_to_cpu_16(rxdp->wb.qword2.ext_status) &
--- 
-1.9.1
-
diff --git a/vnet/vnet/devices/dpdk/device.c b/vnet/vnet/devices/dpdk/device.c
index 20c8b8f77cc..9ea944a2024 100644
--- a/vnet/vnet/devices/dpdk/device.c
+++ b/vnet/vnet/devices/dpdk/device.c
@@ -1136,7 +1136,10 @@ dpdk_subif_add_del_function (vnet_main_t * vnm,
   dpdk_device_t * xd = vec_elt_at_index (xm->devices, hw->dev_instance);
   vnet_sw_interface_t * t = (vnet_sw_interface_t *) st;
   int r, vlan_offload;
+  u32 prev_subifs = xd->vlan_subifs;
 
+  if (is_add) xd->vlan_subifs++;
+  else if (xd->vlan_subifs) xd->vlan_subifs--;
 
   if (xd->dev_type != VNET_DPDK_DEV_ETH)
         return 0;
@@ -1149,21 +1152,26 @@ dpdk_subif_add_del_function (vnet_main_t * vnm,
   if (t->sub.eth.flags.no_tags == 1)
         return 0;
 
-  if ((t->sub.eth.flags.one_tag != 1) || (t->sub.eth.flags.exact_match != 1 ))
+  if ((t->sub.eth.flags.one_tag != 1) || (t->sub.eth.flags.exact_match != 1 )) {
+        xd->vlan_subifs = prev_subifs;
         return clib_error_return (0, "unsupported VLAN setup");
-
+  }
 
   vlan_offload = rte_eth_dev_get_vlan_offload(xd->device_index);
   vlan_offload |= ETH_VLAN_FILTER_OFFLOAD;
 
-  if ((r = rte_eth_dev_set_vlan_offload(xd->device_index, vlan_offload)))
+  if ((r = rte_eth_dev_set_vlan_offload(xd->device_index, vlan_offload))) {
+        xd->vlan_subifs = prev_subifs;
         return clib_error_return (0, "rte_eth_dev_set_vlan_offload[%d]: err %d",
                                   xd->device_index, r);
+  }
 
 
-  if ((r = rte_eth_dev_vlan_filter(xd->device_index, t->sub.eth.outer_vlan_id, is_add)))
+  if ((r = rte_eth_dev_vlan_filter(xd->device_index, t->sub.eth.outer_vlan_id, is_add))) {
+        xd->vlan_subifs = prev_subifs;
         return clib_error_return (0, "rte_eth_dev_vlan_filter[%d]: err %d",
                                  xd->device_index, r);
+  }
 
   return 0;
 }
diff --git a/vnet/vnet/devices/dpdk/dpdk.h b/vnet/vnet/devices/dpdk/dpdk.h
index 019d83f797a..cc22b4390cd 100644
--- a/vnet/vnet/devices/dpdk/dpdk.h
+++ b/vnet/vnet/devices/dpdk/dpdk.h
@@ -208,6 +208,9 @@ typedef struct {
   /* per-worker destination frame queue */
   dpdk_frame_t * frames;
 
+  /* number of sub-interfaces */
+  u16 vlan_subifs;
+
   dpdk_device_type_t dev_type:8;
   dpdk_pmd_t pmd:8;
   i8 cpu_socket;
diff --git a/vnet/vnet/devices/dpdk/node.c b/vnet/vnet/devices/dpdk/node.c
index a7590a5c566..467453ccb37 100644
--- a/vnet/vnet/devices/dpdk/node.c
+++ b/vnet/vnet/devices/dpdk/node.c
@@ -303,9 +303,9 @@ dpdk_rx_next_and_error_from_mb_flags_x1 (dpdk_device_t *xd, struct rte_mbuf *mb,
   else
     {
       *error0 = DPDK_ERROR_NONE;
-      if (xd->per_interface_next_index != ~0)
+      if (PREDICT_FALSE(xd->per_interface_next_index != ~0))
 	n0 = xd->per_interface_next_index;
-      else if (mb_flags & PKT_RX_VLAN_PKT)
+      else if (PREDICT_FALSE(xd->vlan_subifs || (mb_flags & PKT_RX_VLAN_PKT)))
 	n0 = DPDK_RX_NEXT_ETHERNET_INPUT;
       else
 	{