Merge branch 'dpdk_merge' of https://github.com/istokes/ovs into HEAD

Documentation/howto/dpdk.rst

@@ -48,6 +48,18 @@ number of dpdk devices found in the log file::
     $ ovs-vsctl add-port br0 dpdk-p1 -- set Interface dpdk-p1 type=dpdk \
         options:dpdk-devargs=0000:01:00.1
 
+Some NICs (i.e. Mellanox ConnectX-3) have only one PCI address associated
+with multiple ports. Using a PCI device like above won't work. Instead, below
+usage is suggested::
+
+    $ ovs-vsctl add-port br0 dpdk-p0 -- set Interface dpdk-p0 type=dpdk \
+        options:dpdk-devargs="class=eth,mac=00:11:22:33:44:55:01"
+    $ ovs-vsctl add-port br0 dpdk-p1 -- set Interface dpdk-p1 type=dpdk \
+        options:dpdk-devargs="class=eth,mac=00:11:22:33:44:55:02"
+
+Note: such syntax won't support hotplug. The hotplug is supposed to work with
+future DPDK release, v18.05.
+
 After the DPDK ports get added to switch, a polling thread continuously polls
 DPDK devices and consumes 100% of the core, as can be checked from ``top`` and
 ``ps`` commands::

Documentation/intro/install/dpdk.rst

@@ -643,6 +643,18 @@ Limitations
 
 .. _DPDK release notes: http://dpdk.org/doc/guides/rel_notes/release_17_11.html
 
+- Upper bound MTU: DPDK device drivers differ in how the L2 frame for a
+  given MTU value is calculated e.g. i40e driver includes 2 x vlan headers in
+  MTU overhead, em driver includes 1 x vlan header, ixgbe driver does not
+  include a vlan  header in overhead. Currently it is not possible for OVS
+  DPDK to know what upper bound MTU value is supported for a given device.
+  As such OVS DPDK must provision for the case where the L2 frame for a given
+  MTU includes 2 x vlan headers. This reduces the upper bound MTU value for
+  devices that do not include vlan headers in their L2 frames by 8 bytes e.g.
+  ixgbe devices upper bound MTU is reduced from 9710 to 9702. This work
+  around is temporary and is expected to be removed once a method is provided
+  by DPDK to query the upper bound MTU value for a given device.
+
 Reporting Bugs
 --------------
 

lib/dpdk-stub.c

@@ -54,3 +54,8 @@ dpdk_vhost_iommu_enabled(void)
 {
     return false;
 }
+
+void
+print_dpdk_version(void)
+{
+}

lib/dpdk.c

@@ -24,6 +24,7 @@
 
 #include <rte_log.h>
 #include <rte_memzone.h>
+#include <rte_version.h>
 #ifdef DPDK_PDUMP
 #include <rte_mempool.h>
 #include <rte_pdump.h>
@@ -471,6 +472,7 @@ dpdk_init(const struct smap *ovs_other_config)
         static struct ovsthread_once once_enable = OVSTHREAD_ONCE_INITIALIZER;
 
         if (ovsthread_once_start(&once_enable)) {
+            VLOG_INFO("Using %s", rte_version());
             VLOG_INFO("DPDK Enabled - initializing...");
             dpdk_init__(ovs_other_config);
             enabled = true;
@@ -501,3 +503,9 @@ dpdk_set_lcore_id(unsigned cpu)
     ovs_assert(cpu != NON_PMD_CORE_ID);
     RTE_PER_LCORE(_lcore_id) = cpu;
 }
+
+void
+print_dpdk_version(void)
+{
+    puts(rte_version());
+}

lib/dpdk.h

@@ -38,5 +38,6 @@ void dpdk_init(const struct smap *ovs_other_config);
 void dpdk_set_lcore_id(unsigned cpu);
 const char *dpdk_get_vhost_sock_dir(void);
 bool dpdk_vhost_iommu_enabled(void);
+void print_dpdk_version(void);
 
 #endif /* dpdk.h */

lib/netdev-dpdk.c

@@ -452,6 +452,8 @@ struct netdev_rxq_dpdk {
 static void netdev_dpdk_destruct(struct netdev *netdev);
 static void netdev_dpdk_vhost_destruct(struct netdev *netdev);
 
+static void netdev_dpdk_clear_xstats(struct netdev_dpdk *dev);
+
 int netdev_dpdk_get_vid(const struct netdev_dpdk *dev);
 
 struct ingress_policer *
@@ -1104,6 +1106,7 @@ netdev_dpdk_destruct(struct netdev *netdev)
         }
     }
 
+    netdev_dpdk_clear_xstats(dev);
     free(dev->devargs);
     common_destruct(dev);
 
@@ -1169,7 +1172,7 @@ netdev_dpdk_dealloc(struct netdev *netdev)
 }
 
 static void
-netdev_dpdk_clear_xstats(struct netdev_dpdk *dev) OVS_REQUIRES(dev->mutex)
+netdev_dpdk_clear_xstats(struct netdev_dpdk *dev)
 {
     /* If statistics are already allocated, we have to
      * reconfigure, as port_id could have been changed. */
@@ -1274,6 +1277,8 @@ netdev_dpdk_configure_xstats(struct netdev_dpdk *dev)
                     VLOG_WARN("Can't get XSTATS IDs for port: %"PRIu8,
                               dev->port_id);
                 }
+
+                free(rte_xstats);
             }
         }
     } else {
@@ -1336,30 +1341,70 @@ netdev_dpdk_lookup_by_port_id(dpdk_port_t port_id)
     return NULL;
 }
 
+static dpdk_port_t
+netdev_dpdk_get_port_by_mac(const char *mac_str)
+{
+    dpdk_port_t port_id;
+    struct eth_addr mac, port_mac;
+
+    if (!eth_addr_from_string(mac_str, &mac)) {
+        VLOG_ERR("invalid mac: %s", mac_str);
+        return DPDK_ETH_PORT_ID_INVALID;
+    }
+
+    RTE_ETH_FOREACH_DEV (port_id) {
+        struct ether_addr ea;
+
+        rte_eth_macaddr_get(port_id, &ea);
+        memcpy(port_mac.ea, ea.addr_bytes, ETH_ADDR_LEN);
+        if (eth_addr_equals(mac, port_mac)) {
+            return port_id;
+        }
+    }
+
+    return DPDK_ETH_PORT_ID_INVALID;
+}
+
+/*
+ * Normally, a PCI id is enough for identifying a specific DPDK port.
+ * However, for some NICs having multiple ports sharing the same PCI
+ * id, using PCI id won't work then.
+ *
+ * To fix that, here one more method is introduced: "class=eth,mac=$MAC".
+ *
+ * Note that the compatibility is fully kept: user can still use the
+ * PCI id for adding ports (when it's enough for them).
+ */
 static dpdk_port_t
 netdev_dpdk_process_devargs(struct netdev_dpdk *dev,
                             const char *devargs, char **errp)
 {
-    /* Get the name up to the first comma. */
-    char *name = xmemdup0(devargs, strcspn(devargs, ","));
+    char *name;
     dpdk_port_t new_port_id = DPDK_ETH_PORT_ID_INVALID;
 
-    if (rte_eth_dev_get_port_by_name(name, &new_port_id)
-            || !rte_eth_dev_is_valid_port(new_port_id)) {
-        /* Device not found in DPDK, attempt to attach it */
-        if (!rte_eth_dev_attach(devargs, &new_port_id)) {
-            /* Attach successful */
-            dev->attached = true;
-            VLOG_INFO("Device '%s' attached to DPDK", devargs);
-        } else {
-            /* Attach unsuccessful */
-            new_port_id = DPDK_ETH_PORT_ID_INVALID;
-            VLOG_WARN_BUF(errp, "Error attaching device '%s' to DPDK",
-                          devargs);
+    if (strncmp(devargs, "class=eth,mac=", 14) == 0) {
+        new_port_id = netdev_dpdk_get_port_by_mac(&devargs[14]);
+    } else {
+        name = xmemdup0(devargs, strcspn(devargs, ","));
+        if (rte_eth_dev_get_port_by_name(name, &new_port_id)
+                || !rte_eth_dev_is_valid_port(new_port_id)) {
+            /* Device not found in DPDK, attempt to attach it */
+            if (!rte_eth_dev_attach(devargs, &new_port_id)) {
+                /* Attach successful */
+                dev->attached = true;
+                VLOG_INFO("Device '%s' attached to DPDK", devargs);
+            } else {
+                /* Attach unsuccessful */
+                new_port_id = DPDK_ETH_PORT_ID_INVALID;
+            }
         }
+        free(name);
+    }
+
+    if (new_port_id == DPDK_ETH_PORT_ID_INVALID) {
+        VLOG_WARN_BUF(errp, "Error attaching device '%s' to DPDK", devargs);
     }
 
-    free(name);
     return new_port_id;
 }
 
@@ -2134,7 +2179,18 @@ netdev_dpdk_set_mtu(struct netdev *netdev, int mtu)
 {
     struct netdev_dpdk *dev = netdev_dpdk_cast(netdev);
 
-    if (MTU_TO_FRAME_LEN(mtu) > NETDEV_DPDK_MAX_PKT_LEN
+    /* XXX: Ensure that the overall frame length of the requested MTU does not
+     * surpass the NETDEV_DPDK_MAX_PKT_LEN. DPDK device drivers differ in how
+     * the L2 frame length is calculated for a given MTU when
+     * rte_eth_dev_set_mtu(mtu) is called e.g. i40e driver includes 2 x vlan
+     * headers, the em driver includes 1 x vlan header, the ixgbe driver does
+     * not include vlan headers. As such we should use
+     * MTU_TO_MAX_FRAME_LEN(mtu) which includes an additional 2 x vlan headers
+     * (8 bytes) for comparison. This avoids a failure later with
+     * rte_eth_dev_set_mtu(). This approach should be used until DPDK provides
+     * a method to retrieve the upper bound MTU for a given device.
+     */
+    if (MTU_TO_MAX_FRAME_LEN(mtu) > NETDEV_DPDK_MAX_PKT_LEN
         || mtu < ETHER_MIN_MTU) {
         VLOG_WARN("%s: unsupported MTU %d\n", dev->up.name, mtu);
         return EINVAL;
@@ -2359,6 +2415,8 @@ netdev_dpdk_get_custom_stats(const struct netdev *netdev,
              * reconfigured */
             netdev_dpdk_clear_xstats(dev);
         }
+
+        free(values);
     }
 
     ovs_mutex_unlock(&dev->mutex);

utilities/ovs-ctl.in

@@ -72,7 +72,7 @@ set_hostname () {
 set_system_ids () {
     set ovs_vsctl set Open_vSwitch .
 
-    OVS_VERSION=`ovs-vswitchd --version | sed 's/.*) //;1q'`
+    OVS_VERSION=`ovs-vswitchd --version | awk '/Open vSwitch/{print $NF}'`
     set "$@" ovs-version="$OVS_VERSION"
 
     case $SYSTEM_ID in

vswitchd/ovs-vswitchd.c

@@ -187,6 +187,7 @@ parse_options(int argc, char *argv[], char **unixctl_pathp)
 
         case 'V':
             ovs_print_version(0, 0);
+            print_dpdk_version();
             exit(EXIT_SUCCESS);
 
         case OPT_MLOCKALL: