Clean up multi-thread barrier-sync hold-down timer

Main thread: don't bother with the barrier sync hold-down timer if
none of the worker threads are busy.

Worker threads: avoid epoll_pwait (10ms timeout) when the
control-plane has been active in the last half-second.

Cherry-pick a recent dangling reference fix: pool_elt_at_index after
e.g. rx callback is required, in case the unix file pool expands.

Manual feature backport to 18.07

Change-Id: I745fbb8a12aeda34b0ec7b6dcda66c0e25c3eee1
Signed-off-by: Dave Barach <dave@barachs.net>

Makefile

@@ -127,14 +127,14 @@ RPM_SUSE_PLATFORM_DEPS = distribution-release shadow rpm-build
 
 ifeq ($(OS_ID),opensuse)
 ifeq ($(SUSE_NAME),Tumbleweed)
-	RPM_SUSE_DEVEL_DEPS = libboost_headers-devel libboost_thread-devel gcc
+	RPM_SUSE_DEVEL_DEPS = libboost_headers1_68_0-devel-1.68.0  libboost_thread1_68_0-devel-1.68.0 gcc
 	RPM_SUSE_PYTHON_DEPS += python2-ply python2-virtualenv
 endif
 ifeq ($(SUSE_ID),15.0)
-	RPM_SUSE_DEVEL_DEPS = libboost_headers-devel libboost_thread-devel gcc6
+	RPM_SUSE_DEVEL_DEPS = libboost_headers1_68_0-devel-1.68.0  libboost_thread1_68_0-devel-1.68.0 gcc6
 	RPM_SUSE_PYTHON_DEPS += python2-ply python2-virtualenv
 else
-	RPM_SUSE_DEVEL_DEPS += boost_1_61-devel gcc6
+	RPM_SUSE_DEVEL_DEPS += libboost_headers1_68_0-devel-1.68.0 gcc6
 	RPM_SUSE_PYTHON_DEPS += python-virtualenv
 endif
 endif

src/plugins/acl.am

@@ -19,7 +19,7 @@ acl_plugin_la_LDFLAGS += -Wl,-lm,-ldl
 
 
 
-acl_plugin_la_LIBADD =
+acl_plugin_la_DEPENDENCIES =
 
 acl_plugin_la_SOURCES =				\
 	acl/acl.c				\
@@ -48,7 +48,7 @@ libacl_plugin_avx2_la_CFLAGS =                                 \
         $(AM_CFLAGS)  @CPU_AVX2_FLAGS@                          \
         -DCLIB_MARCH_VARIANT=avx2
 noinst_LTLIBRARIES += libacl_plugin_avx2.la
-acl_plugin_la_LIBADD += libacl_plugin_avx2.la
+acl_plugin_la_DEPENDENCIES += libacl_plugin_avx2.la
 endif
 
 if CC_SUPPORTS_AVX512
@@ -60,7 +60,7 @@ libacl_plugin_avx512_la_CFLAGS =                               \
         $(AM_CFLAGS) @CPU_AVX512_FLAGS@                         \
         -DCLIB_MARCH_VARIANT=avx512
 noinst_LTLIBRARIES += libacl_plugin_avx512.la
-acl_plugin_la_LIBADD += libacl_plugin_avx512.la
+acl_plugin_la_DEPENDENCIES += libacl_plugin_avx512.la
 endif
 endif
 

src/plugins/acl/hash_lookup.c

@@ -1478,6 +1478,8 @@ split_partition(acl_main_t *am, u32 first_index,
 	int i=0;
 	u64 collisions = vec_len(pae->colliding_rules);
 	for(i=0; i<collisions; i++){
+                /* reload the hash acl info as it might be a different ACL# */
+	        ha = vec_elt_at_index(am->hash_acl_infos, pae->acl_index);
 
 		DBG( "TM-collision: base_ace:%d (ace_mask:%d, first_collision_mask:%d)",
 				pae->ace_index, pae->mask_type_index, coll_mask_type_index);

src/plugins/avf.am

@@ -14,7 +14,7 @@
 vppplugins_LTLIBRARIES += avf_plugin.la
 vppapitestplugins_LTLIBRARIES += avf_test_plugin.la
 
-avf_plugin_la_LIBADD =
+avf_plugin_la_DEPENDENCIES =
 avf_plugin_la_SOURCES = \
 	avf/cli.c			\
 	avf/device.c			\
@@ -51,7 +51,7 @@ libavf_plugin_avx2_la_CFLAGS =				\
 	$(AM_CFLAGS)  @CPU_AVX2_FLAGS@				\
 	-DCLIB_MARCH_VARIANT=avx2
 noinst_LTLIBRARIES += libavf_plugin_avx2.la
-avf_plugin_la_LIBADD += libavf_plugin_avx2.la
+avf_plugin_la_DEPENDENCIES += libavf_plugin_avx2.la
 endif
 
 if CC_SUPPORTS_AVX512
@@ -63,7 +63,7 @@ libavf_plugin_avx512_la_CFLAGS =				\
 	$(AM_CFLAGS) @CPU_AVX512_FLAGS@				\
 	-DCLIB_MARCH_VARIANT=avx512
 noinst_LTLIBRARIES += libavf_plugin_avx512.la
-avf_plugin_la_LIBADD += libavf_plugin_avx512.la
+avf_plugin_la_DEPENDENCIES += libavf_plugin_avx512.la
 endif
 endif
 

src/plugins/dpdk.am

@@ -31,7 +31,7 @@ dpdk_plugin_la_LDFLAGS += -Wl,-lnuma
 endif
 
 dpdk_plugin_la_LDFLAGS += -Wl,-lm,-ldl
-dpdk_plugin_la_LIBADD =
+dpdk_plugin_la_DEPENDENCIES =
 
 dpdk_plugin_la_SOURCES =					\
   dpdk/main.c							\
@@ -71,7 +71,7 @@ libdpdk_plugin_avx2_la_CFLAGS =					\
 	$(AM_CFLAGS)  @CPU_AVX2_FLAGS@				\
 	-DCLIB_MARCH_VARIANT=avx2
 noinst_LTLIBRARIES += libdpdk_plugin_avx2.la
-dpdk_plugin_la_LIBADD += libdpdk_plugin_avx2.la
+dpdk_plugin_la_DEPENDENCIES += libdpdk_plugin_avx2.la
 endif
 
 if CC_SUPPORTS_AVX512
@@ -83,7 +83,7 @@ libdpdk_plugin_avx512_la_CFLAGS =				\
 	$(AM_CFLAGS) @CPU_AVX512_FLAGS@				\
 	-DCLIB_MARCH_VARIANT=avx512
 noinst_LTLIBRARIES += libdpdk_plugin_avx512.la
-dpdk_plugin_la_LIBADD += libdpdk_plugin_avx512.la
+dpdk_plugin_la_DEPENDENCIES += libdpdk_plugin_avx512.la
 endif
 endif
 

src/plugins/igmp/igmp_report.c

@@ -32,15 +32,21 @@ igmp_group_mk_source_list (const igmp_membership_group_v3_t * r)
   n = clib_net_to_host_u16 (r->n_src_addresses);
 
   if (0 == n)
-    return (NULL);
-
-  vec_validate (srcs, n - 1);
-  s = r->src_addresses;
-
-  for (ii = 0; ii < n; ii++)
     {
-      srcs[ii].ip4 = *s;
+      /* a (*,G) join has no source address specified */
-      s++;
+      vec_validate (srcs, 0);
+      srcs[0].ip4.as_u32 = 0;
+    }
+  else
+    {
+      vec_validate (srcs, n - 1);
+      s = r->src_addresses;
+
+      for (ii = 0; ii < n; ii++)
+	{
+	  srcs[ii].ip4 = *s;
+	  s++;
+	}
     }
 
   return (srcs);

src/plugins/memif.am

@@ -14,7 +14,7 @@
 vppplugins_LTLIBRARIES += memif_plugin.la
 vppapitestplugins_LTLIBRARIES += memif_test_plugin.la
 
-memif_plugin_la_LIBADD =
+memif_plugin_la_DEPENDENCIES =
 memif_plugin_la_SOURCES = memif/memif.c   \
 	memif/memif_api.c                 \
 	memif/cli.c                       \
@@ -49,7 +49,7 @@ memif_plugin_avx2_la_CFLAGS =					\
 	$(AM_CFLAGS)  @CPU_AVX2_FLAGS@				\
 	-DCLIB_MARCH_VARIANT=avx2
 noinst_LTLIBRARIES += memif_plugin_avx2.la
-memif_plugin_la_LIBADD += memif_plugin_avx2.la
+memif_plugin_la_DEPENDENCIES += memif_plugin_avx2.la
 endif
 
 if CC_SUPPORTS_AVX512
@@ -61,7 +61,7 @@ memif_plugin_avx512_la_CFLAGS =					\
 	$(AM_CFLAGS) @CPU_AVX512_FLAGS@				\
 	-DCLIB_MARCH_VARIANT=avx512
 noinst_LTLIBRARIES += memif_plugin_avx512.la
-memif_plugin_la_LIBADD += memif_plugin_avx512.la
+memif_plugin_la_DEPENDENCIES += memif_plugin_avx512.la
 endif
 endif
 

src/vat/api_format.c

@@ -19929,14 +19929,14 @@ vl_api_mpls_fib_path_print (vat_main_t * vam, vl_api_fib_path_t * fp)
     print (vam->ofp,
 	   "  weight %d, sw_if_index %d, is_local %d, is_drop %d, "
 	   "is_unreach %d, is_prohitbit %d, afi %d, next_hop %U",
-	   ntohl (fp->weight), ntohl (fp->sw_if_index), fp->is_local,
+	   fp->weight, ntohl (fp->sw_if_index), fp->is_local,
 	   fp->is_drop, fp->is_unreach, fp->is_prohibit, fp->afi,
 	   format_ip6_address, fp->next_hop);
   else if (fp->afi == IP46_TYPE_IP4)
     print (vam->ofp,
 	   "  weight %d, sw_if_index %d, is_local %d, is_drop %d, "
 	   "is_unreach %d, is_prohitbit %d, afi %d, next_hop %U",
-	   ntohl (fp->weight), ntohl (fp->sw_if_index), fp->is_local,
+	   fp->weight, ntohl (fp->sw_if_index), fp->is_local,
 	   fp->is_drop, fp->is_unreach, fp->is_prohibit, fp->afi,
 	   format_ip4_address, fp->next_hop);
 }

src/vlib/main.h

@@ -61,6 +61,9 @@ typedef struct vlib_main_t
   CLIB_CACHE_LINE_ALIGN_MARK (cacheline0);
   /* Instruction level timing state. */
   clib_time_t clib_time;
+  /* Offset from main thread time */
+  f64 time_offset;
+  f64 time_last_barrier_release;
 
   /* Time stamp of last node dispatch. */
   u64 cpu_time_last_node_dispatch;
@@ -224,7 +227,7 @@ void vlib_worker_loop (vlib_main_t * vm);
 always_inline f64
 vlib_time_now (vlib_main_t * vm)
 {
-  return clib_time_now (&vm->clib_time);
+  return clib_time_now (&vm->clib_time) + vm->time_offset;
 }
 
 always_inline f64

src/vlib/threads.c

@@ -1448,7 +1448,9 @@ vlib_worker_thread_barrier_sync_int (vlib_main_t * vm)
   f64 t_entry;
   f64 t_open;
   f64 t_closed;
+  f64 max_vector_rate;
   u32 count;
+  int i;
 
   if (vec_len (vlib_mains) < 2)
     return;
@@ -1468,13 +1470,43 @@ vlib_worker_thread_barrier_sync_int (vlib_main_t * vm)
       return;
     }
 
+  /*
+   * Need data to decide if we're working hard enough to honor
+   * the barrier hold-down timer.
+   */
+  max_vector_rate = 0.0;
+  for (i = 1; i < vec_len (vlib_mains); i++)
+    max_vector_rate =
+      clib_max (max_vector_rate,
+		vlib_last_vectors_per_main_loop_as_f64 (vlib_mains[i]));
+
   vlib_worker_threads[0].barrier_sync_count++;
 
   /* Enforce minimum barrier open time to minimize packet loss */
   ASSERT (vm->barrier_no_close_before <= (now + BARRIER_MINIMUM_OPEN_LIMIT));
-  while ((now = vlib_time_now (vm)) < vm->barrier_no_close_before)
-    ;
 
+  /*
+   * If any worker thread seems busy, which we define
+   * as a vector rate above 10, we enforce the barrier hold-down timer
+   */
+  if (max_vector_rate > 10.0)
+    {
+      while (1)
+	{
+	  now = vlib_time_now (vm);
+	  /* Barrier hold-down timer expired? */
+	  if (now >= vm->barrier_no_close_before)
+	    break;
+	  if ((vm->barrier_no_close_before - now)
+	      > (2.0 * BARRIER_MINIMUM_OPEN_LIMIT))
+	    {
+	      clib_warning
+		("clock change: would have waited for %.4f seconds",
+		 (vm->barrier_no_close_before - now));
+	      break;
+	    }
+	}
+    }
   /* Record time of closure */
   t_open = now - vm->barrier_epoch;
   vm->barrier_epoch = now;
@@ -1559,6 +1591,14 @@ vlib_worker_thread_barrier_release (vlib_main_t * vm)
 
   deadline = now + BARRIER_SYNC_TIMEOUT;
 
+  /*
+   * Note when we let go of the barrier.
+   * Workers can use this to derive a reasonably accurate
+   * time offset. See vlib_time_now(...)
+   */
+  vm->time_last_barrier_release = vlib_time_now (vm);
+  CLIB_MEMORY_STORE_BARRIER ();
+
   *vlib_worker_threads->wait_at_barrier = 0;
 
   while (*vlib_worker_threads->workers_at_barrier > 0)
@@ -1844,6 +1884,45 @@ threads_init (vlib_main_t * vm)
 
 VLIB_INIT_FUNCTION (threads_init);
 
+
+static clib_error_t *
+show_clock_command_fn (vlib_main_t * vm,
+		       unformat_input_t * input, vlib_cli_command_t * cmd)
+{
+  int i;
+  f64 now;
+
+  now = vlib_time_now (vm);
+
+  vlib_cli_output (vm, "Time now %.9f", now);
+
+  if (vec_len (vlib_mains) == 1)
+    return 0;
+
+  vlib_cli_output (vm, "Time last barrier release %.9f",
+		   vm->time_last_barrier_release);
+
+  for (i = 1; i < vec_len (vlib_mains); i++)
+    {
+      if (vlib_mains[i] == 0)
+	continue;
+      vlib_cli_output (vm, "Thread %d offset %.9f error %.9f", i,
+		       vlib_mains[i]->time_offset,
+		       vm->time_last_barrier_release -
+		       vlib_mains[i]->time_last_barrier_release);
+    }
+  return 0;
+}
+
+/* *INDENT-OFF* */
+VLIB_CLI_COMMAND (f_command, static) =
+{
+  .path = "show clock",
+  .short_help = "show clock",
+  .function = show_clock_command_fn,
+};
+/* *INDENT-ON* */
+
 /*
  * fd.io coding-style-patch-verification: ON
  *

src/vlib/threads.h

@@ -400,7 +400,7 @@ vlib_worker_thread_barrier_check (void)
 {
   if (PREDICT_FALSE (*vlib_worker_threads->wait_at_barrier))
     {
-      vlib_main_t *vm;
+      vlib_main_t *vm = vlib_get_main ();
       clib_smp_atomic_add (vlib_worker_threads->workers_at_barrier, 1);
       if (CLIB_DEBUG > 0)
 	{
@@ -409,6 +409,21 @@ vlib_worker_thread_barrier_check (void)
 	}
       while (*vlib_worker_threads->wait_at_barrier)
 	;
+
+      /*
+       * Recompute the offset from thread-0 time.
+       * Note that vlib_time_now adds vm->time_offset, so
+       * clear it first. Save the resulting idea of "now", to
+       * see how well we're doing. See show_clock_command_fn(...)
+       */
+      {
+	f64 now;
+	vm->time_offset = 0.0;
+	now = vlib_time_now (vm);
+	vm->time_offset = vlib_global_main.time_last_barrier_release - now;
+	vm->time_last_barrier_release = vlib_time_now (vm);
+      }
+
       if (CLIB_DEBUG > 0)
 	vm->parked_at_barrier = 0;
       clib_smp_atomic_add (vlib_worker_threads->workers_at_barrier, -1);

src/vlib/unix/input.c

@@ -145,9 +145,13 @@ linux_epoll_input_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
     vlib_node_main_t *nm = &vm->node_main;
     u32 ticks_until_expiration;
     f64 timeout;
+    f64 now;
     int timeout_ms = 0, max_timeout_ms = 10;
     f64 vector_rate = vlib_last_vectors_per_main_loop (vm);
 
+    if (is_main == 0)
+      now = vlib_time_now (vm);
+
     /*
      * If we've been asked for a fixed-sleep between main loop polls,
      * do so right away.
@@ -194,8 +198,9 @@ linux_epoll_input_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
 	  }
 	node->input_main_loops_per_call = 0;
       }
-    else if (is_main == 0 && vector_rate < 2 &&
+    else if (is_main == 0 && vector_rate < 2
-	     nm->input_node_counts_by_state[VLIB_NODE_STATE_POLLING] == 0)
+	     && (vlib_global_main.time_last_barrier_release + 0.5 < now)
+	     && nm->input_node_counts_by_state[VLIB_NODE_STATE_POLLING] == 0)
       {
 	timeout = 10e-3;
 	timeout_ms = max_timeout_ms;
@@ -223,12 +228,32 @@ linux_epoll_input_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
 				      em->epoll_events,
 				      vec_len (em->epoll_events), timeout_ms);
 	  }
+
       }
     else
       {
+	/*
+	 * Worker thread, no epoll fd's, sleep for 100us at a time
+	 * and check for a barrier sync request
+	 */
 	if (timeout_ms)
-	  usleep (timeout_ms * 1000);
+	  {
-	return 0;
+	    struct timespec ts, tsrem;
+	    f64 limit = now + (f64) timeout_ms * 1e-3;
+
+	    while (vlib_time_now (vm) < limit)
+	      {
+		/* Sleep for 100us at a time */
+		ts.tv_sec = 0;
+		ts.tv_nsec = 1000 * 100;
+
+		while (nanosleep (&ts, &tsrem) < 0)
+		  ts = tsrem;
+		if (*vlib_worker_threads->wait_at_barrier)
+		  goto done;
+	      }
+	  }
+	goto done;
       }
   }
 
@@ -238,7 +263,7 @@ linux_epoll_input_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
 	vlib_panic_with_error (vm, clib_error_return_unix (0, "epoll_wait"));
 
       /* non fatal error (e.g. EINTR). */
-      return 0;
+      goto done;
     }
 
   em->epoll_waits += 1;
@@ -314,6 +339,7 @@ linux_epoll_input_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
 	}
     }
 
+done:
   return 0;
 }
 

src/vnet.am

@@ -1294,7 +1294,7 @@ libvnet_avx2_la_CFLAGS =			\
 	$(AM_CFLAGS)  @CPU_AVX2_FLAGS@		\
 	-DCLIB_MARCH_VARIANT=avx2
 noinst_LTLIBRARIES += libvnet_avx2.la
-libvnet_la_LIBADD += libvnet_avx2.la
+libvnet_la_DEPENDENCIES += libvnet_avx2.la
 endif
 
 if CC_SUPPORTS_AVX512
@@ -1306,7 +1306,7 @@ libvnet_avx512_la_CFLAGS =			\
 	$(AM_CFLAGS) @CPU_AVX512_FLAGS@		\
 	-DCLIB_MARCH_VARIANT=avx512
 noinst_LTLIBRARIES += libvnet_avx512.la
-libvnet_la_LIBADD += libvnet_avx512.la
+libvnet_la_DEPENDENCIES += libvnet_avx512.la
 endif
 endif
 

src/vnet/bier/bier_disp_dispatch_node.c

@@ -72,6 +72,7 @@ bier_disp_dispatch_inline (vlib_main_t * vm,
             bdei0 = vnet_buffer(b0)->ip.adj_index[VLIB_TX];
             hdr0 = vlib_buffer_get_current(b0);
             bde0 = bier_disp_entry_get(bdei0);
+            vnet_buffer(b0)->ip.adj_index[VLIB_RX] = BIER_RX_ITF;
 
             /*
              * header is in network order - flip it, we are about to

src/vnet/bier/bier_drop.c

@@ -87,11 +87,10 @@ format_bier_drop_trace (u8 * s, va_list * args)
   return s;
 }
 
-/* *INDENT-OFF* */
 VLIB_REGISTER_NODE (bier_drop_node, static) =
 {
-    .function = bier_drop,.
+    .function = bier_drop,
-    name = "bier-drop",
+    .name = "bier-drop",
     .vector_size = sizeof (u32),
     .format_trace = format_bier_drop_trace,
     .n_next_nodes = 1,

src/vnet/bier/bier_hdr_inlines.h

@@ -26,6 +26,11 @@
 #include <vnet/bier/bier_bit_string.h>
 #include <vnet/ip/ip6_packet.h>
 
+/**
+ * Special Value of the BIER RX interface
+ */
+#define BIER_RX_ITF (~0 - 1)
+
 /**
  * Mask and shift values for the fields incorporated
  * into the header's first word

src/vnet/bier/bier_imp.c

@@ -68,7 +68,7 @@ bier_imp_add_or_lock (const bier_table_id_t *bti,
     pool_get_aligned(bier_imp_pool, bi, CLIB_CACHE_LINE_BYTES);
 
     bi->bi_tbl = *bti;
-    btii = bier_table_add_or_lock(bti, MPLS_LABEL_INVALID);
+    btii = bier_table_lock(bti);
 
     /*
      * init the BIER header we will paint on in the data plane

src/vnet/bier/bier_imp_node.c

@@ -120,30 +120,39 @@ bier_imp_dpo_inline (vlib_main_t * vm,
             vlib_buffer_advance(b0, -(sizeof(bier_hdr_t) +
                                       bier_hdr_len_id_to_num_bytes(bimp0->bi_tbl.bti_hdr_len)));
             hdr0 = vlib_buffer_get_current(b0);
-            clib_memcpy(hdr0, &bimp0->bi_hdr,
-                        (sizeof(bier_hdr_t) +
-                         bier_hdr_len_id_to_num_bytes(bimp0->bi_tbl.bti_hdr_len)));
-            /*
-             * Fixup the entropy and protocol, both of which have a
-             * zero value post the paint job
-             */
-            hdr0->bh_oam_dscp_proto |=
-                clib_host_to_net_u16(bproto << BIER_HDR_PROTO_FIELD_SHIFT);
-            hdr0->bh_first_word |=
-                clib_host_to_net_u32((vnet_buffer(b0)->ip.flow_hash &
-                                      BIER_HDR_ENTROPY_FIELD_MASK) <<
-                                     BIER_HDR_ENTROPY_FIELD_SHIFT);
 
-            /*
+            /* RPF check */
-             * use TTL 64 for the post enacp MPLS label/BIFT-ID
+            if (PREDICT_FALSE(BIER_RX_ITF == vnet_buffer(b0)->ip.adj_index[VLIB_RX]))
-             * this we be decremeted in bier_output node.
+            {
-             */
+                next0 = 0;
-            vnet_buffer(b0)->mpls.ttl = 65;
+            }
+            else
+            {
+                clib_memcpy(hdr0, &bimp0->bi_hdr,
+                            (sizeof(bier_hdr_t) +
+                             bier_hdr_len_id_to_num_bytes(bimp0->bi_tbl.bti_hdr_len)));
+                /*
+                 * Fixup the entropy and protocol, both of which have a
+                 * zero value post the paint job
+                 */
+                hdr0->bh_oam_dscp_proto |=
+                    clib_host_to_net_u16(bproto << BIER_HDR_PROTO_FIELD_SHIFT);
+                hdr0->bh_first_word |=
+                    clib_host_to_net_u32((vnet_buffer(b0)->ip.flow_hash &
+                                          BIER_HDR_ENTROPY_FIELD_MASK) <<
+                                         BIER_HDR_ENTROPY_FIELD_SHIFT);
 
-            /* next node */
+                /*
-            next0 = bimp0->bi_dpo[fproto].dpoi_next_node;
+                 * use TTL 64 for the post enacp MPLS label/BIFT-ID
-            vnet_buffer(b0)->ip.adj_index[VLIB_TX] =
+                 * this we be decremeted in bier_output node.
-                bimp0->bi_dpo[fproto].dpoi_index;
+                 */
+                vnet_buffer(b0)->mpls.ttl = 65;
+
+                /* next node */
+                next0 = bimp0->bi_dpo[fproto].dpoi_next_node;
+                vnet_buffer(b0)->ip.adj_index[VLIB_TX] =
+                    bimp0->bi_dpo[fproto].dpoi_index;
+            }
 
             if (PREDICT_FALSE(b0->flags & VLIB_BUFFER_IS_TRACED))
             {
@@ -194,7 +203,7 @@ VLIB_REGISTER_NODE (bier_imp_ip4_node) = {
     .format_trace = format_bier_imp_trace,
     .n_next_nodes = 1,
     .next_nodes = {
-        [0] = "error-drop",
+        [0] = "bier-drop",
     }
 };
 VLIB_NODE_FUNCTION_MULTIARCH (bier_imp_ip4_node, bier_imp_ip4)

src/vnet/bier/bier_table.c

@@ -329,6 +329,70 @@ bier_table_mk_ecmp (index_t bti)
     return (bt);
 }
 
+
+static index_t
+bier_table_create (const bier_table_id_t *btid,
+                   mpls_label_t local_label)
+{
+    /*
+     * add a new table
+     */
+    bier_table_t *bt;
+    index_t bti;
+    u32 key;
+
+    key = bier_table_mk_key(btid);
+
+    pool_get_aligned(bier_table_pool, bt, CLIB_CACHE_LINE_BYTES);
+    bier_table_init(bt, btid, local_label);
+
+    hash_set(bier_tables_by_key, key, bier_table_get_index(bt));
+    bti = bier_table_get_index(bt);
+
+    if (bier_table_is_main(bt))
+    {
+        bt = bier_table_mk_ecmp(bti);
+
+        /*
+         * add whichever mpls-fib or bift we need
+         */
+        if (local_label != MPLS_LABEL_INVALID)
+        {
+            bt->bt_ll = local_label;
+            bier_table_mk_lfib(bt);
+        }
+        else
+        {
+            bier_table_mk_bift(bt);
+        }
+    }
+
+    return (bti);
+}
+
+index_t
+bier_table_lock (const bier_table_id_t *btid)
+{
+    bier_table_t *bt;
+    index_t bti;
+
+    bt = bier_table_find(btid);
+
+    if (NULL == bt)
+    {
+        bti = bier_table_create(btid, MPLS_LABEL_INVALID);
+        bt = bier_table_get(bti);
+    }
+    else
+    {
+        bti = bier_table_get_index(bt);
+    }
+
+    bier_table_lock_i(bt);
+
+    return (bti);
+}
+
 index_t
 bier_table_add_or_lock (const bier_table_id_t *btid,
                         mpls_label_t local_label)
@@ -379,36 +443,8 @@ bier_table_add_or_lock (const bier_table_id_t *btid,
     }
     else
     {
-        /*
+        bti = bier_table_create(btid, local_label);
-         * add a new table
+        bt = bier_table_get(bti);
-         */
-        u32 key;
-
-        key = bier_table_mk_key(btid);
-
-        pool_get_aligned(bier_table_pool, bt, CLIB_CACHE_LINE_BYTES);
-        bier_table_init(bt, btid, local_label);
-
-        hash_set(bier_tables_by_key, key, bier_table_get_index(bt));
-        bti = bier_table_get_index(bt);
-
-        if (bier_table_is_main(bt))
-        {
-            bt = bier_table_mk_ecmp(bti);
-
-            /*
-             * add whichever mpls-fib or bift we need
-             */
-            if (local_label != MPLS_LABEL_INVALID)
-            {
-                bt->bt_ll = local_label;
-                bier_table_mk_lfib(bt);
-            }
-            else
-            {
-                bier_table_mk_bift(bt);
-            }
-        }
     }
 
     bier_table_lock_i(bt);

src/vnet/bier/bier_table.h

@@ -91,6 +91,7 @@ STATIC_ASSERT((sizeof(bier_table_t) <= 2*CLIB_CACHE_LINE_BYTES),
 
 extern index_t bier_table_add_or_lock(const bier_table_id_t *id,
                                       mpls_label_t ll);
+extern index_t bier_table_lock(const bier_table_id_t *id);
 extern void bier_table_unlock(const bier_table_id_t *id);
 
 extern void bier_table_route_add(const bier_table_id_t *bti,

src/vnet/bonding/cli.c

@@ -516,11 +516,13 @@ bond_enslave (vlib_main_t * vm, bond_enslave_args_t * args)
       ethernet_set_rx_redirect (vnm, sif_hw, 1);
     }
 
-  if ((bif->mode == BOND_MODE_LACP) && bm->lacp_enable_disable)
+  if (bif->mode == BOND_MODE_LACP)
     {
-      (*bm->lacp_enable_disable) (vm, bif, sif, 1);
+      if (bm->lacp_enable_disable)
+	(*bm->lacp_enable_disable) (vm, bif, sif, 1);
     }
-  else
+  else if (sif->port_enabled &&
+	   (sif_hw->flags & VNET_HW_INTERFACE_FLAG_LINK_UP))
     {
       bond_enable_collecting_distributing (vm, sif);
     }

src/vnet/bonding/node.c

@@ -404,19 +404,21 @@ bond_sw_interface_up_down (vnet_main_t * vnm, u32 sw_if_index, u32 flags)
   if (sif)
     {
       sif->port_enabled = flags & VNET_SW_INTERFACE_FLAG_ADMIN_UP;
+      if (sif->lacp_enabled)
+	return 0;
+
       if (sif->port_enabled == 0)
 	{
-	  if (sif->lacp_enabled == 0)
+	  bond_disable_collecting_distributing (vm, sif);
-	    {
-	      bond_disable_collecting_distributing (vm, sif);
-	    }
 	}
       else
 	{
-	  if (sif->lacp_enabled == 0)
+	  vnet_main_t *vnm = vnet_get_main ();
-	    {
+	  vnet_hw_interface_t *hw =
-	      bond_enable_collecting_distributing (vm, sif);
+	    vnet_get_sup_hw_interface (vnm, sw_if_index);
-	    }
+
+	  if (hw->flags & VNET_HW_INTERFACE_FLAG_LINK_UP)
+	    bond_enable_collecting_distributing (vm, sif);
 	}
     }
 
@@ -437,19 +439,16 @@ bond_hw_interface_up_down (vnet_main_t * vnm, u32 hw_if_index, u32 flags)
   sif = bond_get_slave_by_sw_if_index (sw->sw_if_index);
   if (sif)
     {
+      if (sif->lacp_enabled)
+	return 0;
+
       if (!(flags & VNET_HW_INTERFACE_FLAG_LINK_UP))
 	{
-	  if (sif->lacp_enabled == 0)
+	  bond_disable_collecting_distributing (vm, sif);
-	    {
-	      bond_disable_collecting_distributing (vm, sif);
-	    }
 	}
-      else
+      else if (sif->port_enabled)
 	{
-	  if (sif->lacp_enabled == 0)
+	  bond_enable_collecting_distributing (vm, sif);
-	    {
-	      bond_enable_collecting_distributing (vm, sif);
-	    }
 	}
     }
 

src/vnet/devices/virtio/vhost_user.h

@@ -245,6 +245,14 @@ typedef struct
 
   /* The rx queue policy (interrupt/adaptive/polling) for this queue */
   u32 mode;
+
+  /*
+   * It contains the device queue number. -1 if it does not. The idea is
+   * to not invoke vnet_hw_interface_assign_rx_thread and
+   * vnet_hw_interface_unassign_rx_thread more than once for the duration of
+   * the interface even if it is disconnected and reconnected.
+   */
+  i16 qid;
 } vhost_user_vring_t;
 
 #define VHOST_USER_EVENT_START_TIMER 1
@@ -288,9 +296,6 @@ typedef struct
   /* Whether to use spinlock or per_cpu_tx_qid assignment */
   u8 use_tx_spinlock;
   u16 *per_cpu_tx_qid;
-
-  /* Vector of active rx queues for this interface */
-  u16 *rx_queues;
 } vhost_user_intf_t;
 
 typedef struct

src/vnet/devices/virtio/vhost_user_api.c

@@ -234,6 +234,9 @@ vhost_user_api_hookup (vlib_main_t * vm)
   foreach_vpe_api_msg;
 #undef _
 
+  /* Mark CREATE_VHOST_USER_IF as mp safe */
+  am->is_mp_safe[VL_API_CREATE_VHOST_USER_IF] = 1;
+
   /*
    * Set up the (msg_name, crc, message-id) table
    */

src/vnet/fib/fib_types.c

@@ -305,6 +305,22 @@ fib_forw_chain_type_from_dpo_proto (dpo_proto_t proto)
     return (FIB_FORW_CHAIN_TYPE_UNICAST_IP4);
 }
 
+fib_forward_chain_type_t
+fib_forw_chain_type_from_fib_proto (fib_protocol_t proto)
+{
+    switch (proto)
+    {
+    case FIB_PROTOCOL_IP4:
+	return (FIB_FORW_CHAIN_TYPE_UNICAST_IP4);
+    case FIB_PROTOCOL_IP6:
+	return (FIB_FORW_CHAIN_TYPE_UNICAST_IP6);
+    case FIB_PROTOCOL_MPLS:
+	return (FIB_FORW_CHAIN_TYPE_MPLS_NON_EOS);
+    }
+    ASSERT(0);
+    return (FIB_FORW_CHAIN_TYPE_UNICAST_IP4);
+}
+
 vnet_link_t
 fib_forw_chain_type_to_link_type (fib_forward_chain_type_t fct)
 {

src/vnet/fib/fib_types.h

@@ -177,6 +177,11 @@ extern fib_forward_chain_type_t fib_forw_chain_type_from_link_type(vnet_link_t l
  */
 extern fib_forward_chain_type_t fib_forw_chain_type_from_dpo_proto(dpo_proto_t proto);
 
+/**
+ * @brief Convert from a fib-protocol to a chain type.
+ */
+extern fib_forward_chain_type_t fib_forw_chain_type_from_fib_proto(fib_protocol_t proto);
+
 /**
  * @brief Convert from a chain type to the DPO proto it will install
  */

src/vnet/gre/interface.c

@@ -127,7 +127,9 @@ gre_tunnel_from_fib_node (fib_node_t * node)
 void
 gre_tunnel_stack (adj_index_t ai)
 {
+  fib_forward_chain_type_t fib_fwd;
   gre_main_t *gm = &gre_main;
+  dpo_id_t tmp = DPO_INVALID;
   ip_adjacency_t *adj;
   gre_tunnel_t *gt;
   u32 sw_if_index;
@@ -149,9 +151,7 @@ gre_tunnel_stack (adj_index_t ai)
       return;
     }
 
-  dpo_id_t tmp = DPO_INVALID;
+  fib_fwd = fib_forw_chain_type_from_fib_proto (gt->tunnel_dst.fp_proto);
-  fib_forward_chain_type_t fib_fwd = (FIB_PROTOCOL_IP6 == adj->ia_nh_proto) ?
-    FIB_FORW_CHAIN_TYPE_UNICAST_IP6 : FIB_FORW_CHAIN_TYPE_UNICAST_IP4;
 
   fib_entry_contribute_forwarding (gt->fib_entry_index, fib_fwd, &tmp);
   if (DPO_LOAD_BALANCE == tmp.dpoi_type)

src/vnet/interface_api.c

@@ -1223,6 +1223,11 @@ interface_api_hookup (vlib_main_t * vm)
   foreach_vpe_api_msg;
 #undef _
 
+  /* Mark these APIs as mp safe */
+  am->is_mp_safe[VL_API_SW_INTERFACE_DUMP] = 1;
+  am->is_mp_safe[VL_API_SW_INTERFACE_DETAILS] = 1;
+  am->is_mp_safe[VL_API_SW_INTERFACE_TAG_ADD_DEL] = 1;
+
   /*
    * Set up the (msg_name, crc, message-id) table
    */

src/vnet/interface_cli.c

@@ -469,6 +469,7 @@ VLIB_CLI_COMMAND (show_sw_interfaces_command, static) = {
   .path = "show interface",
   .short_help = "show interface [address|addr|features|feat] [<interface> [<interface> [..]]] [verbose]",
   .function = show_sw_interfaces,
+  .is_mp_safe = 1,
 };
 /* *INDENT-ON* */