vnet: distinguish between max_frame_size and MTU

Type: improvement
Change-Id: I3659de6599f402c92e3855e3bf0e5e3388f2bea0
Signed-off-by: Damjan Marion <damarion@cisco.com>

17 files changed, 148 insertions, 106 deletions

diff --git a/src/plugins/af_xdp/device.c b/src/plugins/af_xdp/device.c
index a5e0b739300..87d1251c229 100644
--- a/src/plugins/af_xdp/device.c
+++ b/src/plugins/af_xdp/device.c
@@ -68,7 +68,8 @@ af_xdp_mac_change (vnet_hw_interface_t * hw, const u8 * old, const u8 * new)
 }
 
 static clib_error_t *
-af_xdp_set_mtu (vnet_main_t *vnm, vnet_hw_interface_t *hw, u32 mtu)
+af_xdp_set_max_frame_size (vnet_main_t *vnm, vnet_hw_interface_t *hw,
+			   u32 frame_size)
 {
   af_xdp_main_t *am = &af_xdp_main;
   af_xdp_device_t *ad = vec_elt_at_index (am->devices, hw->dev_instance);
@@ -657,7 +658,7 @@ af_xdp_create_if (vlib_main_t * vm, af_xdp_create_if_args_t * args)
   eir.dev_instance = ad->dev_instance;
   eir.address = ad->hwaddr;
   eir.cb.flag_change = af_xdp_flag_change;
-  eir.cb.set_mtu = af_xdp_set_mtu;
+  eir.cb.set_max_frame_size = af_xdp_set_max_frame_size;
   ad->hw_if_index = vnet_eth_register_interface (vnm, &eir);
 
   sw = vnet_get_hw_sw_interface (vnm, ad->hw_if_index);
diff --git a/src/plugins/dpdk/device/common.c b/src/plugins/dpdk/device/common.c
index 3eca847de79..0f54a616773 100644
--- a/src/plugins/dpdk/device/common.c
+++ b/src/plugins/dpdk/device/common.c
@@ -61,6 +61,7 @@ dpdk_device_setup (dpdk_device_t * xd)
   vlib_thread_main_t *tm = vlib_get_thread_main ();
   vnet_sw_interface_t *sw = vnet_get_sw_interface (vnm, xd->sw_if_index);
   vnet_hw_interface_t *hi = vnet_get_hw_interface (vnm, xd->hw_if_index);
+  u16 buf_sz = vlib_buffer_get_default_data_size (vm);
   vnet_hw_if_caps_change_t caps = {};
   struct rte_eth_dev_info dev_info;
   struct rte_eth_conf conf = {};
@@ -161,49 +162,51 @@ dpdk_device_setup (dpdk_device_t * xd)
 
 #if RTE_VERSION < RTE_VERSION_NUM(21, 11, 0, 0)
   if (rxo & DEV_RX_OFFLOAD_JUMBO_FRAME)
-    conf.rxmode.max_rx_pkt_len =
-      clib_min (ETHERNET_MAX_PACKET_BYTES, dev_info.max_rx_pktlen);
+    {
+      conf.rxmode.max_rx_pkt_len = dev_info.max_rx_pktlen;
+      xd->max_supported_frame_size = dev_info.max_rx_pktlen;
+      mtu = xd->max_supported_frame_size - xd->driver_frame_overhead;
+    }
+  else
+    {
+      mtu = 1500;
+      xd->max_supported_frame_size = mtu + xd->driver_frame_overhead;
+    }
 #else
-  dpdk_log_debug ("[%u] min_mtu: %u, max_mtu: %u, min_rx_bufsize: %u, "
-		  "max_rx_pktlen: %u, max_lro_pkt_size: %u",
-		  xd->port_id, dev_info.min_mtu, dev_info.max_mtu,
-		  dev_info.min_rx_bufsize, dev_info.max_rx_pktlen,
-		  dev_info.max_lro_pkt_size);
-
-  mtu = xd->conf.disable_multi_seg ? 2000 : ETHERNET_MAX_PACKET_BYTES;
-  conf.rxmode.mtu = clib_min (mtu, dev_info.max_rx_pktlen);
+  if (xd->conf.disable_multi_seg)
+    xd->max_supported_frame_size = clib_min (dev_info.max_rx_pktlen, buf_sz);
+  else
+    xd->max_supported_frame_size = dev_info.max_rx_pktlen;
+#endif
+
+  mtu = clib_min (xd->max_supported_frame_size - xd->driver_frame_overhead,
+		  ethernet_main.default_mtu);
+  mtu = mtu + hi->frame_overhead - xd->driver_frame_overhead;
+
+#if RTE_VERSION >= RTE_VERSION_NUM(21, 11, 0, 0)
+  conf.rxmode.mtu = mtu;
 #endif
 
 retry:
   rv = rte_eth_dev_configure (xd->port_id, xd->conf.n_rx_queues,
 			      xd->conf.n_tx_queues, &conf);
-
   if (rv < 0 && conf.intr_conf.rxq)
     {
       conf.intr_conf.rxq = 0;
       goto retry;
     }
 
-  if (rv < 0)
-    {
-      dpdk_device_error (xd, "rte_eth_dev_configure", rv);
-      goto error;
-    }
-
-  rte_eth_dev_get_mtu (xd->port_id, &mtu);
-  dpdk_log_debug ("[%u] device default mtu %u", xd->port_id, mtu);
+#if RTE_VERSION < RTE_VERSION_NUM(21, 11, 0, 0)
+  rte_eth_dev_set_mtu (xd->port_id, mtu);
+#endif
 
-  hi->max_supported_packet_bytes = mtu;
-  if (hi->max_packet_bytes > mtu)
-    {
-      vnet_hw_interface_set_mtu (vnm, xd->hw_if_index, mtu);
-    }
-  else
-    {
-      rte_eth_dev_set_mtu (xd->port_id, hi->max_packet_bytes);
-      dpdk_log_debug ("[%u] port mtu set to %u", xd->port_id,
-		      hi->max_packet_bytes);
-    }
+  hi->max_frame_size = 0;
+  vnet_hw_interface_set_max_frame_size (vnm, xd->hw_if_index,
+					mtu + hi->frame_overhead);
+  dpdk_log_debug ("[%u] mtu %u max_frame_size %u max max_frame_size %u "
+		  "driver_frame_overhead %u",
+		  xd->port_id, mtu, hi->max_frame_size,
+		  xd->max_supported_frame_size, xd->driver_frame_overhead);
 
   vec_validate_aligned (xd->tx_queues, xd->conf.n_tx_queues - 1,
 			CLIB_CACHE_LINE_BYTES);
diff --git a/src/plugins/dpdk/device/dpdk.h b/src/plugins/dpdk/device/dpdk.h
index f1cd17da1cd..196f68f9756 100644
--- a/src/plugins/dpdk/device/dpdk.h
+++ b/src/plugins/dpdk/device/dpdk.h
@@ -200,6 +200,13 @@ typedef struct
   /* mac address */
   u8 *default_mac_address;
 
+  /* maximum supported max frame size */
+  u16 max_supported_frame_size;
+
+  /* due to lack of API to get ethernet max_frame_size we store information
+   * deducted from device info */
+  u8 driver_frame_overhead;
+
   /* error string */
   clib_error_t *errors;
   dpdk_port_conf_t conf;
diff --git a/src/plugins/dpdk/device/format.c b/src/plugins/dpdk/device/format.c
index 4a4b4110b9f..4ef22729532 100644
--- a/src/plugins/dpdk/device/format.c
+++ b/src/plugins/dpdk/device/format.c
@@ -199,8 +199,8 @@ format_dpdk_link_status (u8 * s, va_list * args)
 
       s = format (s, "%s duplex ", (l->link_duplex == ETH_LINK_FULL_DUPLEX) ?
 		  "full" : "half");
-      s = format (s, "mtu %d %s\n", hi->max_packet_bytes, promisc ?
-		  " promisc" : "");
+      s = format (s, "max-frame-size %d %s\n", hi->max_frame_size,
+		  promisc ? " promisc" : "");
     }
   else
     s = format (s, "\n");
diff --git a/src/plugins/dpdk/device/init.c b/src/plugins/dpdk/device/init.c
index 4faf6290dba..096b8d382fe 100644
--- a/src/plugins/dpdk/device/init.c
+++ b/src/plugins/dpdk/device/init.c
@@ -70,11 +70,15 @@ const struct
 };
 
 static clib_error_t *
-dpdk_set_mtu (vnet_main_t *vnm, vnet_hw_interface_t *hi, u32 mtu)
+dpdk_set_max_frame_size (vnet_main_t *vnm, vnet_hw_interface_t *hi,
+			 u32 frame_size)
 {
   dpdk_main_t *dm = &dpdk_main;
   dpdk_device_t *xd = vec_elt_at_index (dm->devices, hi->dev_instance);
   int rv;
+  u32 mtu;
+
+  mtu = frame_size - xd->driver_frame_overhead;
 
   rv = rte_eth_dev_set_mtu (xd->port_id, mtu);
 
@@ -99,7 +103,8 @@ dpdk_set_mtu (vnet_main_t *vnm, vnet_hw_interface_t *hi, u32 mtu)
 	}
     }
   else
-    dpdk_log_debug ("[%u] mtu set to %u", xd->port_id, mtu);
+    dpdk_log_debug ("[%u] max_frame_size set to %u by setting MTU to %u",
+		    xd->port_id, frame_size, mtu);
 
   return 0;
 }
@@ -376,6 +381,21 @@ dpdk_lib_init (dpdk_main_t * dm)
 	  xd->conf.rss_hf &= di.flow_type_rss_offloads;
 	}
 
+      xd->driver_frame_overhead =
+	RTE_ETHER_HDR_LEN + 2 * RTE_VLAN_HLEN + RTE_ETHER_CRC_LEN;
+#if RTE_VERSION >= RTE_VERSION_NUM(21, 11, 0, 0)
+      q = di.max_rx_pktlen - di.max_mtu;
+
+      if (q < xd->driver_frame_overhead && q > 0)
+	xd->driver_frame_overhead = q;
+      dpdk_log_debug ("[%u] min_mtu: %u, max_mtu: %u, min_rx_bufsize: %u, "
+		      "max_rx_pktlen: %u, max_lro_pkt_size: %u",
+		      xd->port_id, di.min_mtu, di.max_mtu, di.min_rx_bufsize,
+		      di.max_rx_pktlen, di.max_lro_pkt_size);
+#endif
+      dpdk_log_debug ("[%u] driver frame overhead is %u", port_id,
+		      xd->driver_frame_overhead);
+
       /* number of RX and TX tescriptors */
       if (devconf->num_rx_desc)
 	xd->conf.n_rx_desc = devconf->num_rx_desc;
@@ -397,7 +417,7 @@ dpdk_lib_init (dpdk_main_t * dm)
       eir.dev_instance = xd->device_index;
       eir.address = addr;
       eir.cb.flag_change = dpdk_flag_change;
-      eir.cb.set_mtu = dpdk_set_mtu;
+      eir.cb.set_max_frame_size = dpdk_set_max_frame_size;
       xd->hw_if_index = vnet_eth_register_interface (vnm, &eir);
       hi = vnet_get_hw_interface (vnm, xd->hw_if_index);
       hi->numa_node = xd->cpu_socket = (i8) rte_eth_dev_socket_id (port_id);
diff --git a/src/plugins/rdma/device.c b/src/plugins/rdma/device.c
index 167a23213cc..3fdc399a4e3 100644
--- a/src/plugins/rdma/device.c
+++ b/src/plugins/rdma/device.c
@@ -184,7 +184,8 @@ rdma_mac_change (vnet_hw_interface_t * hw, const u8 * old, const u8 * new)
 }
 
 static clib_error_t *
-rdma_set_mtu (vnet_main_t *vnm, vnet_hw_interface_t *hw, u32 mtu)
+rdma_set_max_frame_size (vnet_main_t *vnm, vnet_hw_interface_t *hw,
+			 u32 frame_size)
 {
   return vnet_error (VNET_ERR_UNSUPPORTED, 0);
 }
@@ -358,7 +359,7 @@ rdma_register_interface (vnet_main_t * vnm, rdma_device_t * rd)
   eir.dev_instance = rd->dev_instance;
   eir.address = rd->hwaddr.bytes;
   eir.cb.flag_change = rdma_flag_change;
-  eir.cb.set_mtu = rdma_set_mtu;
+  eir.cb.set_max_frame_size = rdma_set_max_frame_size;
   rd->hw_if_index = vnet_eth_register_interface (vnm, &eir);
   /* Indicate ability to support L3 DMAC filtering and
    * initialize interface to L3 non-promisc mode */
diff --git a/src/vnet/devices/af_packet/af_packet.c b/src/vnet/devices/af_packet/af_packet.c
index cf4f91e2fcd..208537996e3 100644
--- a/src/vnet/devices/af_packet/af_packet.c
+++ b/src/vnet/devices/af_packet/af_packet.c
@@ -57,13 +57,15 @@ unsigned int if_nametoindex (const char *ifname);
 typedef struct tpacket_req tpacket_req_t;
 
 static clib_error_t *
-af_packet_eth_set_mtu (vnet_main_t *vnm, vnet_hw_interface_t *hi, u32 mtu)
+af_packet_eth_set_max_frame_size (vnet_main_t *vnm, vnet_hw_interface_t *hi,
+				  u32 frame_size)
 {
   clib_error_t *error, *rv;
   af_packet_main_t *apm = &af_packet_main;
   af_packet_if_t *apif = pool_elt_at_index (apm->interfaces, hi->dev_instance);
 
-  error = vnet_netlink_set_link_mtu (apif->host_if_index, mtu);
+  error = vnet_netlink_set_link_mtu (apif->host_if_index,
+				     frame_size + hi->frame_overhead);
 
   if (error)
     {
@@ -75,7 +77,7 @@ af_packet_eth_set_mtu (vnet_main_t *vnm, vnet_hw_interface_t *hi, u32 mtu)
       return rv;
     }
   else
-    apif->host_mtu = mtu;
+    apif->host_mtu = frame_size + hi->frame_overhead;
   return 0;
 }
 
@@ -398,7 +400,7 @@ af_packet_create_if (af_packet_create_if_arg_t *arg)
       eir.dev_class_index = af_packet_device_class.index;
       eir.dev_instance = if_index;
       eir.address = hw_addr;
-      eir.cb.set_mtu = af_packet_eth_set_mtu;
+      eir.cb.set_max_frame_size = af_packet_eth_set_max_frame_size;
       apif->hw_if_index = vnet_eth_register_interface (vnm, &eir);
     }
   else
diff --git a/src/vnet/devices/tap/tap.c b/src/vnet/devices/tap/tap.c
index 2e7ad39f3b4..f5ed30ad7b0 100644
--- a/src/vnet/devices/tap/tap.c
+++ b/src/vnet/devices/tap/tap.c
@@ -683,12 +683,10 @@ tap_create_if (vlib_main_t * vm, tap_create_if_args_t * args)
     }
   if (vif->type == VIRTIO_IF_TYPE_TUN)
     {
-      hw->max_supported_packet_bytes = TUN_MAX_PACKET_BYTES;
-      hw->min_packet_bytes = hw->min_supported_packet_bytes =
-	TUN_MIN_PACKET_BYTES;
-      hw->max_packet_bytes =
-	args->host_mtu_size ? args->host_mtu_size : TUN_DEFAULT_PACKET_BYTES;
-      vnet_sw_interface_set_mtu (vnm, hw->sw_if_index, hw->max_packet_bytes);
+      hw->min_frame_size = TUN_MIN_PACKET_BYTES;
+      vnet_hw_interface_set_mtu (
+	vnm, hw->sw_if_index,
+	args->host_mtu_size ? args->host_mtu_size : TUN_DEFAULT_PACKET_BYTES);
     }
 
   vnet_hw_if_change_caps (vnm, vif->hw_if_index, &cc);
diff --git a/src/vnet/ethernet/ethernet.h b/src/vnet/ethernet/ethernet.h
index b6adeb6f44d..858400d08d8 100644
--- a/src/vnet/ethernet/ethernet.h
+++ b/src/vnet/ethernet/ethernet.h
@@ -133,8 +133,8 @@ typedef struct
   /* ethernet interface flags change */
   ethernet_flag_change_function_t *flag_change;
 
-  /* set MTU callback */
-  vnet_interface_set_mtu_function_t *set_mtu;
+  /* set Max Frame Size callback */
+  vnet_interface_set_max_frame_size_function_t *set_max_frame_size;
 } vnet_eth_if_callbacks_t;
 
 #define ETHERNET_MIN_PACKET_BYTES  64
@@ -576,6 +576,8 @@ typedef struct
 {
   u32 dev_class_index;
   u32 dev_instance;
+  u16 max_frame_size;
+  u16 frame_overhead;
   vnet_eth_if_callbacks_t cb;
   const u8 *address;
 } vnet_eth_interface_registration_t;
diff --git a/src/vnet/ethernet/interface.c b/src/vnet/ethernet/interface.c
index bac882228cb..a0ad7989867 100644
--- a/src/vnet/ethernet/interface.c
+++ b/src/vnet/ethernet/interface.c
@@ -311,15 +311,18 @@ ethernet_mac_change (vnet_hw_interface_t * hi,
 }
 
 static clib_error_t *
-ethernet_set_mtu (vnet_main_t *vnm, vnet_hw_interface_t *hi, u32 mtu)
+ethernet_set_max_frame_size (vnet_main_t *vnm, vnet_hw_interface_t *hi,
+			     u32 frame_size)
 {
   ethernet_interface_t *ei =
     pool_elt_at_index (ethernet_main.interfaces, hi->hw_instance);
 
-  if (ei->cb.set_mtu)
-    return ei->cb.set_mtu (vnm, hi, mtu);
+  if (ei->cb.set_max_frame_size)
+    return ei->cb.set_max_frame_size (vnm, hi, frame_size);
 
-  return 0;
+  return vnet_error (
+    VNET_ERR_UNSUPPORTED,
+    "underlying driver doesn't support changing Max Frame Size");
 }
 
 /* *INDENT-OFF* */
@@ -333,7 +336,7 @@ VNET_HW_INTERFACE_CLASS (ethernet_hw_interface_class) = {
   .build_rewrite = ethernet_build_rewrite,
   .update_adjacency = ethernet_update_adjacency,
   .mac_addr_change_function = ethernet_mac_change,
-  .set_mtu = ethernet_set_mtu,
+  .set_max_frame_size = ethernet_set_max_frame_size,
 };
 /* *INDENT-ON* */
 
@@ -378,10 +381,15 @@ vnet_eth_register_interface (vnet_main_t *vnm,
 
   ethernet_setup_node (vnm->vlib_main, hi->output_node_index);
 
-  hi->min_packet_bytes = hi->min_supported_packet_bytes =
-    ETHERNET_MIN_PACKET_BYTES;
-  hi->max_supported_packet_bytes = ETHERNET_MAX_PACKET_BYTES;
-  hi->max_packet_bytes = em->default_mtu;
+  hi->min_frame_size = ETHERNET_MIN_PACKET_BYTES;
+  hi->frame_overhead =
+    r->frame_overhead ?
+      r->max_frame_size :
+      sizeof (ethernet_header_t) + 2 * sizeof (ethernet_vlan_header_t);
+  hi->max_frame_size = r->max_frame_size ?
+			 r->max_frame_size :
+			 ethernet_main.default_mtu + hi->frame_overhead;
+  ;
 
   /* Default ethernet MTU, 9000 unless set by ethernet_config see below */
   vnet_sw_interface_set_mtu (vnm, hi->sw_if_index, em->default_mtu);
diff --git a/src/vnet/gre/interface.c b/src/vnet/gre/interface.c
index bc78c605068..bb0be865664 100644
--- a/src/vnet/gre/interface.c
+++ b/src/vnet/gre/interface.c
@@ -447,13 +447,13 @@ vnet_gre_tunnel_add (vnet_gre_tunnel_add_del_args_t * a,
 
   if (!is_ipv6)
     {
-      hi->min_packet_bytes =
-	64 + sizeof (gre_header_t) + sizeof (ip4_header_t);
+      hi->frame_overhead = sizeof (gre_header_t) + sizeof (ip4_header_t);
+      hi->min_frame_size = hi->frame_overhead + 64;
     }
   else
     {
-      hi->min_packet_bytes =
-	64 + sizeof (gre_header_t) + sizeof (ip6_header_t);
+      hi->frame_overhead = sizeof (gre_header_t) + sizeof (ip6_header_t);
+      hi->min_frame_size = hi->frame_overhead + 64;
     }
 
   /* Standard default gre MTU. */
diff --git a/src/vnet/interface.c b/src/vnet/interface.c
index a1493c66c23..412c6755ea8 100644
--- a/src/vnet/interface.c
+++ b/src/vnet/interface.c
@@ -769,30 +769,41 @@ sw_interface_walk_callback (vnet_main_t * vnm, u32 sw_if_index, void *ctx)
 }
 
 clib_error_t *
-vnet_hw_interface_set_mtu (vnet_main_t *vnm, u32 hw_if_index, u32 mtu)
+vnet_hw_interface_set_max_frame_size (vnet_main_t *vnm, u32 hw_if_index,
+				      u32 fs)
 {
   vnet_hw_interface_t *hi = vnet_get_hw_interface (vnm, hw_if_index);
   vnet_hw_interface_class_t *hw_if_class =
     vnet_get_hw_interface_class (vnm, hi->hw_class_index);
   clib_error_t *err = 0;
 
-  if (hi->max_packet_bytes != mtu)
+  log_debug ("set_max_frame_size: interface %s, max_frame_size %u -> %u",
+	     hi->name, hi->max_frame_size, fs);
+
+  if (hw_if_class->set_max_frame_size == 0)
+    return vnet_error (VNET_ERR_UNSUPPORTED,
+		       "hw class doesn't support changing Max Frame Size");
+
+  if (hi->max_frame_size != fs)
     {
-      if (mtu > hi->max_supported_packet_bytes ||
-	  mtu < hi->min_supported_packet_bytes)
-	return vnet_error (VNET_ERR_INVALID_VALUE,
-			   "requested mtu must be in the %u to %u range",
-			   hi->min_supported_packet_bytes,
-			   hi->max_supported_packet_bytes);
-      if (hw_if_class->set_mtu)
-	if ((err = hw_if_class->set_mtu (vnm, hi, mtu)))
+      u32 mtu;
+      if (hw_if_class->set_max_frame_size)
+	if ((err = hw_if_class->set_max_frame_size (vnm, hi, fs)))
 	  return err;
-      hi->max_packet_bytes = mtu;
+      hi->max_frame_size = fs;
+      mtu = fs - hi->frame_overhead;
       vnet_hw_interface_walk_sw (vnm, hw_if_index, sw_interface_walk_callback,
 				 &mtu);
     }
   return 0;
 }
+clib_error_t *
+vnet_hw_interface_set_mtu (vnet_main_t *vnm, u32 hw_if_index, u32 mtu)
+{
+  vnet_hw_interface_t *hi = vnet_get_hw_interface (vnm, hw_if_index);
+  return vnet_hw_interface_set_max_frame_size (vnm, hw_if_index,
+					       mtu + hi->frame_overhead);
+}
 
 static void
 setup_tx_node (vlib_main_t * vm,
@@ -910,7 +921,6 @@ vnet_register_interface (vnet_main_t * vnm,
   hw->hw_instance = hw_instance;
 
   hw->max_rate_bits_per_sec = 0;
-  hw->min_packet_bytes = 0;
   vnet_sw_interface_set_mtu (vnm, hw->sw_if_index, 0);
 
   if (dev_class->tx_function == 0 && dev_class->tx_fn_registrations == 0)
diff --git a/src/vnet/interface.h b/src/vnet/interface.h
index 2eb50aed5e8..c8fbc61ec7b 100644
--- a/src/vnet/interface.h
+++ b/src/vnet/interface.h
@@ -71,7 +71,7 @@ typedef clib_error_t *(vnet_subif_add_del_function_t)
    struct vnet_sw_interface_t * template, int is_add);
 
 /* Interface set mtu callback. */
-typedef clib_error_t *(vnet_interface_set_mtu_function_t) (
+typedef clib_error_t *(vnet_interface_set_max_frame_size_function_t) (
   struct vnet_main_t *vnm, struct vnet_hw_interface_t *hi, u32 mtu);
 
 /* Interface set mac address callback. */
@@ -432,8 +432,9 @@ typedef struct _vnet_hw_interface_class
 
   /* Function to add/delete additional MAC addresses */
   vnet_interface_add_del_mac_address_function_t *mac_addr_add_del_function;
-  /* Function to set mtu. */
-  vnet_interface_set_mtu_function_t *set_mtu;
+
+  /* Function to set max frame size. */
+  vnet_interface_set_max_frame_size_function_t *set_max_frame_size;
 
   /* Format function to display interface name. */
   format_function_t *format_interface_name;
@@ -701,17 +702,14 @@ typedef struct vnet_hw_interface_t
   /* Maximum transmit rate for this interface in bits/sec. */
   f64 max_rate_bits_per_sec;
 
-  /* Smallest packet size supported by this interface. */
-  u32 min_supported_packet_bytes;
-
-  /* Largest packet size supported by this interface. */
-  u32 max_supported_packet_bytes;
-
   /* Smallest packet size for this interface. */
-  u32 min_packet_bytes;
+  u32 min_frame_size;
+
+  /* Largest frame size for this interface. */
+  u32 max_frame_size;
 
-  /* Largest packet size for this interface. */
-  u32 max_packet_bytes;
+  /* Layer 2 overhead */
+  u16 frame_overhead;
 
   /* Hash table mapping sub interface id to sw_if_index. */
   uword *sub_interface_sw_if_index_by_id;
diff --git a/src/vnet/interface_api.c b/src/vnet/interface_api.c
index d70cd1e42f8..938f3bb327e 100644
--- a/src/vnet/interface_api.c
+++ b/src/vnet/interface_api.c
@@ -258,7 +258,7 @@ send_sw_interface_details (vpe_api_main_t * am,
   mp->link_duplex = ntohl (((hi->flags & VNET_HW_INTERFACE_FLAG_DUPLEX_MASK) >>
 			    VNET_HW_INTERFACE_FLAG_DUPLEX_SHIFT));
   mp->link_speed = ntohl (hi->link_speed);
-  mp->link_mtu = ntohs (hi->max_packet_bytes);
+  mp->link_mtu = ntohs (hi->max_frame_size - hi->frame_overhead);
   mp->mtu[VNET_MTU_L3] = ntohl (swif->mtu[VNET_MTU_L3]);
   mp->mtu[VNET_MTU_IP4] = ntohl (swif->mtu[VNET_MTU_IP4]);
   mp->mtu[VNET_MTU_IP6] = ntohl (swif->mtu[VNET_MTU_IP6]);
diff --git a/src/vnet/interface_cli.c b/src/vnet/interface_cli.c
index 740a8af78ec..b6d38578714 100644
--- a/src/vnet/interface_cli.c
+++ b/src/vnet/interface_cli.c
@@ -1167,21 +1167,11 @@ mtu_cmd (vlib_main_t * vm, unformat_input_t * input, vlib_cli_command_t * cmd)
        * Change physical MTU on interface. Only supported for Ethernet
        * interfaces
        */
-      vnet_hw_interface_t *hi = vnet_get_hw_interface (vnm, hw_if_index);
       ethernet_interface_t *eif = ethernet_get_interface (em, hw_if_index);
 
       if (!eif)
 	return clib_error_return (0, "not supported");
 
-      if (mtu < hi->min_supported_packet_bytes)
-	return clib_error_return (0, "Invalid mtu (%d): "
-				  "must be >= min pkt bytes (%d)", mtu,
-				  hi->min_supported_packet_bytes);
-
-      if (mtu > hi->max_supported_packet_bytes)
-	return clib_error_return (0, "Invalid mtu (%d): must be <= (%d)", mtu,
-				  hi->max_supported_packet_bytes);
-
       err = vnet_hw_interface_set_mtu (vnm, hw_if_index, mtu);
       if (err)
 	return err;
diff --git a/src/vnet/interface_funcs.h b/src/vnet/interface_funcs.h
index 57e0d33b406..5434542cd7c 100644
--- a/src/vnet/interface_funcs.h
+++ b/src/vnet/interface_funcs.h
@@ -310,7 +310,7 @@ always_inline u32
 vnet_hw_interface_get_mtu (vnet_main_t * vnm, u32 hw_if_index)
 {
   vnet_hw_interface_t *hw = vnet_get_hw_interface (vnm, hw_if_index);
-  return hw->max_packet_bytes;
+  return hw->max_frame_size - hw->frame_overhead;
 }
 
 always_inline u32
@@ -430,6 +430,10 @@ clib_error_t *set_hw_interface_rx_placement (u32 hw_if_index, u32 queue_id,
 /* Set tx-queue placement on the interface */
 int set_hw_interface_tx_queue (u32 hw_if_index, u32 queue_id, uword *bitmap);
 
+/* Set the Max Frame Size on the HW interface */
+clib_error_t *vnet_hw_interface_set_max_frame_size (vnet_main_t *vnm,
+						    u32 hw_if_index,
+						    u32 max_frame_size);
 /* Set the MTU on the HW interface */
 clib_error_t *vnet_hw_interface_set_mtu (vnet_main_t *vnm, u32 hw_if_index,
 					 u32 mtu);
diff --git a/src/vnet/ipip/ipip.c b/src/vnet/ipip/ipip.c
index 2ac234eb7ca..600f5421125 100644
--- a/src/vnet/ipip/ipip.c
+++ b/src/vnet/ipip/ipip.c
@@ -782,13 +782,11 @@ ipip_add_tunnel (ipip_transport_t transport,
   gm->tunnel_index_by_sw_if_index[sw_if_index] = t_idx;
 
   if (t->transport == IPIP_TRANSPORT_IP4)
-    {
-      hi->min_packet_bytes = 64 + sizeof (ip4_header_t);
-    }
+    hi->frame_overhead = sizeof (ip4_header_t);
   else
-    {
-      hi->min_packet_bytes = 64 + sizeof (ip6_header_t);
-    }
+    hi->frame_overhead = sizeof (ip6_header_t);
+
+  hi->min_frame_size = hi->frame_overhead + 64;
 
   /* Standard default ipip MTU. */
   vnet_sw_interface_set_mtu (vnm, sw_if_index, 9000);