netdev-offload: Add multi-thread API.

Expose functions reporting user configuration of offloading threads, as
well as utility functions for multithreading.

This will only expose the configuration knob to the user, while no
datapath will implement the multiple thread request.

This will allow implementations to use this API for offload thread
management in relevant layers before enabling the actual dataplane
implementation.

The offload thread ID is lazily allocated and can as such be in a
different order than the offload thread start sequence.

The RCU thread will sometime access hardware-offload objects from
a provider for reclamation purposes.  In such case, it will get
a default offload thread ID of 0. Care must be taken that using
this thread ID is safe concurrently with the offload threads.

Signed-off-by: Gaetan Rivet <grive@u256.net>
Reviewed-by: Eli Britstein <elibr@nvidia.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Signed-off-by: Ilya Maximets <i.maximets@ovn.org>

NEWS

@@ -7,6 +7,8 @@ Post-v2.16.0
        to 'hash', enables hash-based Tx packet steering mode to utilize all the
        Tx queues of the interface regardles of the number of PMD threads.
      * Removed experimental tag for PMD Auto Load Balance.
+     * New configuration knob 'other_config:n-offload-threads' to change the
+       number of HW offloading threads.
    - DPDK:
      * EAL argument --socket-mem is no longer configured by default upon
        start-up.  If dpdk-socket-mem and dpdk-alloc-mem are not specified,

lib/netdev-offload-provider.h

@@ -84,6 +84,7 @@ struct netdev_flow_api {
                     struct dpif_flow_stats *);
 
     /* Get the number of flows offloaded to netdev.
+     * 'n_flows' is an array of counters, one per offload thread.
      * Return 0 if successful, otherwise returns a positive errno value. */
     int (*flow_get_n_flows)(struct netdev *, uint64_t *n_flows);
 

lib/netdev-offload.c

@@ -60,6 +60,12 @@ VLOG_DEFINE_THIS_MODULE(netdev_offload);
 
 static bool netdev_flow_api_enabled = false;
 
+#define DEFAULT_OFFLOAD_THREAD_NB 1
+#define MAX_OFFLOAD_THREAD_NB 10
+
+static unsigned int offload_thread_nb = DEFAULT_OFFLOAD_THREAD_NB;
+DEFINE_EXTERN_PER_THREAD_DATA(netdev_offload_thread_id, OVSTHREAD_ID_UNSET);
+
 /* Protects 'netdev_flow_apis'.  */
 static struct ovs_mutex netdev_flow_api_provider_mutex = OVS_MUTEX_INITIALIZER;
 
@@ -448,6 +454,64 @@ netdev_is_flow_api_enabled(void)
     return netdev_flow_api_enabled;
 }
 
+unsigned int
+netdev_offload_thread_nb(void)
+{
+    return offload_thread_nb;
+}
+
+unsigned int
+netdev_offload_ufid_to_thread_id(const ovs_u128 ufid)
+{
+    uint32_t ufid_hash;
+
+    if (netdev_offload_thread_nb() == 1) {
+        return 0;
+    }
+
+    ufid_hash = hash_words64_inline(
+            (const uint64_t [2]){ ufid.u64.lo,
+                                  ufid.u64.hi }, 2, 1);
+    return ufid_hash % netdev_offload_thread_nb();
+}
+
+unsigned int
+netdev_offload_thread_init(void)
+{
+    static atomic_count next_id = ATOMIC_COUNT_INIT(0);
+    bool thread_is_hw_offload;
+    bool thread_is_rcu;
+
+    thread_is_hw_offload = !strncmp(get_subprogram_name(),
+                                    "hw_offload", strlen("hw_offload"));
+    thread_is_rcu = !strncmp(get_subprogram_name(), "urcu", strlen("urcu"));
+
+    /* Panic if any other thread besides offload and RCU tries
+     * to initialize their thread ID. */
+    ovs_assert(thread_is_hw_offload || thread_is_rcu);
+
+    if (*netdev_offload_thread_id_get() == OVSTHREAD_ID_UNSET) {
+        unsigned int id;
+
+        if (thread_is_rcu) {
+            /* RCU will compete with other threads for shared object access.
+             * Reclamation functions using a thread ID must be thread-safe.
+             * For that end, and because RCU must consider all potential shared
+             * objects anyway, its thread-id can be whichever, so return 0.
+             */
+            id = 0;
+        } else {
+            /* Only the actual offload threads have their own ID. */
+            id = atomic_count_inc(&next_id);
+        }
+        /* Panic if any offload thread is getting a spurious ID. */
+        ovs_assert(id < netdev_offload_thread_nb());
+        return *netdev_offload_thread_id_get() = id;
+    } else {
+        return *netdev_offload_thread_id_get();
+    }
+}
+
 void
 netdev_ports_flow_flush(const char *dpif_type)
 {
@@ -660,7 +724,16 @@ netdev_ports_get_n_flows(const char *dpif_type, odp_port_t port_no,
     ovs_rwlock_rdlock(&netdev_hmap_rwlock);
     data = netdev_ports_lookup(port_no, dpif_type);
     if (data) {
-        ret = netdev_flow_get_n_flows(data->netdev, n_flows);
+        uint64_t thread_n_flows[MAX_OFFLOAD_THREAD_NB] = {0};
+        unsigned int tid;
+
+        ret = netdev_flow_get_n_flows(data->netdev, thread_n_flows);
+        *n_flows = 0;
+        if (!ret) {
+            for (tid = 0; tid < netdev_offload_thread_nb(); tid++) {
+                *n_flows += thread_n_flows[tid];
+            }
+        }
     }
     ovs_rwlock_unlock(&netdev_hmap_rwlock);
     return ret;
@@ -713,7 +786,22 @@ netdev_set_flow_api_enabled(const struct smap *ovs_other_config)
         if (ovsthread_once_start(&once)) {
             netdev_flow_api_enabled = true;
 
-            VLOG_INFO("netdev: Flow API Enabled");
+            offload_thread_nb = smap_get_ullong(ovs_other_config,
+                                                "n-offload-threads",
+                                                DEFAULT_OFFLOAD_THREAD_NB);
+            if (offload_thread_nb > MAX_OFFLOAD_THREAD_NB) {
+                VLOG_WARN("netdev: Invalid number of threads requested: %u",
+                          offload_thread_nb);
+                offload_thread_nb = DEFAULT_OFFLOAD_THREAD_NB;
+            }
+
+            if (smap_get(ovs_other_config, "n-offload-threads")) {
+                VLOG_INFO("netdev: Flow API Enabled, using %u thread%s",
+                          offload_thread_nb,
+                          offload_thread_nb > 1 ? "s" : "");
+            } else {
+                VLOG_INFO("netdev: Flow API Enabled");
+            }
 
 #ifdef __linux__
             tc_set_policy(smap_get_def(ovs_other_config, "tc-policy",

lib/netdev-offload.h

@@ -21,6 +21,7 @@
 #include "openvswitch/netdev.h"
 #include "openvswitch/types.h"
 #include "ovs-rcu.h"
+#include "ovs-thread.h"
 #include "packets.h"
 #include "flow.h"
 
@@ -81,6 +82,24 @@ struct offload_info {
     odp_port_t orig_in_port; /* Originating in_port for tnl flows. */
 };
 
+DECLARE_EXTERN_PER_THREAD_DATA(unsigned int, netdev_offload_thread_id);
+
+unsigned int netdev_offload_thread_nb(void);
+unsigned int netdev_offload_thread_init(void);
+unsigned int netdev_offload_ufid_to_thread_id(const ovs_u128 ufid);
+
+static inline unsigned int
+netdev_offload_thread_id(void)
+{
+    unsigned int id = *netdev_offload_thread_id_get();
+
+    if (OVS_UNLIKELY(id == OVSTHREAD_ID_UNSET)) {
+        id = netdev_offload_thread_init();
+    }
+
+    return id;
+}
+
 int netdev_flow_flush(struct netdev *);
 int netdev_flow_dump_create(struct netdev *, struct netdev_flow_dump **dump,
                             bool terse);

vswitchd/vswitch.xml

@@ -247,6 +247,22 @@
         </p>
       </column>
 
+      <column name="other_config" key="n-offload-threads"
+              type='{"type": "integer", "minInteger": 1, "maxInteger": 10}'>
+        <p>
+          Set this value to the number of threads created to manage hardware
+          offloads.
+        </p>
+        <p>
+          The default value is <code>1</code>. Changing this value requires
+          restarting the daemon.
+        </p>
+        <p>
+          This is only relevant for userspace datapath and only if
+          <ref column="other_config" key="hw-offload"/> is enabled.
+        </p>
+      </column>
+
       <column name="other_config" key="tc-policy"
               type='{"type": "string",
                      "enum": ["set", ["none", "skip_sw", "skip_hw"]]}'>