vnet: distinguish between max_frame_size and MTU

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

4 files changed, 65 insertions, 35 deletions

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);