assign flood_class to vnet_sw_interface_t template in subif api handle function

Change-Id: I352f4a4adcf8771c21530657efcaecb532416612
Signed-off-by: Joe Zhou <zhouyoucn@qq.com>
(cherry picked from commit 715f94ed94638ea883f919361bff7a3f46fd1d1b)

.gitreview

@@ -2,3 +2,4 @@
 host=gerrit.fd.io
 port=29418
 project=vpp
+defaultbranch=stable/1901

Makefile

@@ -133,8 +133,7 @@ ifeq ($(SUSE_NAME),Tumbleweed)
 	RPM_SUSE_PYTHON_DEPS += python2-ply python2-virtualenv
 endif
 ifeq ($(SUSE_ID),15.0)
-	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
+	RPM_SUSE_DEVEL_DEPS = libboost_headers-devel libboost_thread-devel gcc6
 else
 	RPM_SUSE_DEVEL_DEPS += libboost_headers1_68_0-devel-1.68.0 gcc6
 	RPM_SUSE_PYTHON_DEPS += python-virtualenv
@@ -143,8 +142,8 @@ endif
 
 ifeq ($(OS_ID),opensuse-leap)
 ifeq ($(SUSE_ID),15.0)
-	RPM_SUSE_DEVEL_DEPS = libboost_headers-devel libboost_thread-devel gcc6
-	RPM_SUSE_PYTHON_DEPS += python2-ply python2-virtualenv
+	RPM_SUSE_DEVEL_DEPS = libboost_headers-devel libboost_thread-devel gcc
+	RPM_SUSE_PYTHON_DEPS += python3-ply python2-virtualenv
 endif
 endif
 
@@ -295,6 +294,9 @@ endif
 else ifneq ("$(wildcard /etc/redhat-release)","")
 ifeq ($(OS_ID),rhel)
 	@sudo -E yum-config-manager --enable rhel-server-rhscl-7-rpms
+	@sudo -E yum groupinstall $(CONFIRM) $(RPM_DEPENDS_GROUPS)
+	@sudo -E yum install $(CONFIRM) $(RPM_DEPENDS)
+	@sudo -E debuginfo-install $(CONFIRM) glibc openssl-libs mbedtls-devel zlib
 else ifeq ($(OS_ID),centos)
 	@sudo -E yum install $(CONFIRM) centos-release-scl-rh
 	@sudo -E yum groupinstall $(CONFIRM) $(RPM_DEPENDS_GROUPS)

doxygen/test_framework_doc.md

@@ -4,6 +4,7 @@ Test Framework Documentation    {#test_framework_doc}
 PyDoc generated documentation for the "make test" framework is available for
 the following releases:
 
+- [Test framework documentation for VPP 19.01](https://docs.fd.io/vpp/19.01/vpp_make_test/html)
 - [Test framework documentation for VPP 18.10](https://docs.fd.io/vpp/18.10/vpp_make_test/html)
 - [Test framework documentation for VPP 18.07](https://docs.fd.io/vpp/18.07/vpp_make_test/html)
 - [Test framework documentation for VPP 18.04](https://docs.fd.io/vpp/18.04/vpp_make_test/html)

extras/scripts/list_api_changes.py

@@ -1,8 +1,8 @@
 #!/usr/bin/env python
 import os, fnmatch, subprocess
 
-starttag = 'v18.10-rc0'
-endtag = 'v18.10'
+starttag = 'v19.01-rc0'
+endtag = 'v19.01'
 emit_md = True
 apifiles = []
 

extras/vom/vom/route.cpp

@@ -632,7 +632,8 @@ ip_mroute::dump(std::ostream& os)
 ip_mroute::event_handler::event_handler()
 {
   OM::register_listener(this);
-  inspect::register_handler({ "ip-route" }, "ip route configurations", this);
+  inspect::register_handler({ "ip-mroute" },
+                            "ip multicast route configurations", this);
 }
 
 void

src/CMakeLists.txt

@@ -87,16 +87,6 @@ foreach(DIR ${SUBDIRS})
   add_subdirectory(${DIR})
 endforeach()
 
-##############################################################################
-# packaging
-##############################################################################
-include(cmake/pack.cmake)
-add_vpp_packaging(
-  NAME "vpp"
-  VENDOR "fd.io"
-  DESCRIPTION "Vector Packet Processor"
-)
-
 ##############################################################################
 # detect if we are inside git repo and add configure dependency
 ##############################################################################
@@ -105,6 +95,7 @@ execute_process(
   COMMAND git rev-parse --show-toplevel
   OUTPUT_VARIABLE VPP_GIT_TOPLEVEL_DIR
   OUTPUT_STRIP_TRAILING_WHITESPACE
+  ERROR_QUIET
 )
 
 if (VPP_GIT_TOPLEVEL_DIR)

src/plugins/avf/device.c

@@ -653,14 +653,19 @@ clib_error_t *
 avf_op_enable_queues (vlib_main_t * vm, avf_device_t * ad, u32 rx, u32 tx)
 {
   virtchnl_queue_select_t qs = { 0 };
-  int i;
+  int i = 0;
   qs.vsi_id = ad->vsi_id;
   qs.rx_queues = rx;
   qs.tx_queues = tx;
-  for (i = 0; i < ad->n_rx_queues; i++)
+  while (rx)
     {
-      avf_rxq_t *rxq = vec_elt_at_index (ad->rxqs, i);
-      avf_reg_write (ad, AVF_QRX_TAIL (i), rxq->n_enqueued);
+      if (rx & (1 << i))
+	{
+	  avf_rxq_t *rxq = vec_elt_at_index (ad->rxqs, i);
+	  avf_reg_write (ad, AVF_QRX_TAIL (i), rxq->n_enqueued);
+	  rx &= ~(1 << i);
+	}
+      i++;
     }
   return avf_send_to_pf (vm, ad, VIRTCHNL_OP_ENABLE_QUEUES, &qs,
 			 sizeof (virtchnl_queue_select_t), 0, 0);
@@ -850,10 +855,8 @@ avf_device_init (vlib_main_t * vm, avf_main_t * am, avf_device_t * ad,
   if ((error = avf_op_add_eth_addr (vm, ad, 1, ad->hwaddr)))
     return error;
 
-  if ((error = avf_op_enable_queues (vm, ad, ad->n_rx_queues, 0)))
-    return error;
-
-  if ((error = avf_op_enable_queues (vm, ad, 0, ad->n_tx_queues)))
+  if ((error = avf_op_enable_queues (vm, ad, pow2_mask (ad->n_rx_queues),
+				     pow2_mask (ad->n_tx_queues))))
     return error;
 
   ad->flags |= AVF_DEVICE_F_INITIALIZED;

src/plugins/avf/input.c

@@ -411,6 +411,7 @@ no_more_desc:
 
       if ((or_qw1 & AVF_RXD_ERROR_IPE) == 0)
 	f->flags |= ETH_INPUT_FRAME_F_IP4_CKSUM_OK;
+      vlib_frame_no_append (f);
     }
 
   n_left_to_next -= n_rx_packets;

src/plugins/avf/output.c

@@ -45,6 +45,9 @@ avf_tx_enqueue (vlib_main_t * vm, avf_txq_t * txq, u32 * buffers,
   /* avoid ring wrap */
   n_desc_left = txq->size - clib_max (txq->next, txq->n_enqueued + 8);
 
+  if (n_desc_left == 0)
+    return 0;
+
   while (n_packets_left && n_desc_left)
     {
       u32 or_flags;
@@ -140,7 +143,7 @@ VNET_DEVICE_CLASS_TX_FN (avf_device_class) (vlib_main_t * vm,
   avf_txq_t *txq = vec_elt_at_index (ad->txqs, qid % ad->num_queue_pairs);
   u32 *buffers = vlib_frame_vector_args (frame);
   u16 n_enq, n_left;
-  u16 n_retry = 5;
+  u16 n_retry = 2;
 
   clib_spinlock_lock_if_init (&txq->lock);
 
@@ -158,10 +161,11 @@ retry:
 	  if (slot == 0)
 	    break;
 
-	  complete_slot = slot[0];
-	  if (avf_tx_desc_get_dtyp (txq->descs + complete_slot) != 0x0F)
+	  if (avf_tx_desc_get_dtyp (txq->descs + slot[0]) != 0x0F)
 	    break;
 
+	  complete_slot = slot[0];
+
 	  clib_ring_deq (txq->rs_slots);
 	}
 

src/plugins/dpdk/device/node.c

@@ -389,6 +389,7 @@ dpdk_device_input (vlib_main_t * vm, dpdk_main_t * dm, dpdk_device_t * xd,
 	  if (xd->flags & DPDK_DEVICE_FLAG_RX_IP4_CKSUM &&
 	      (or_flags & PKT_RX_IP_CKSUM_BAD) == 0)
 	    f->flags |= ETH_INPUT_FRAME_F_IP4_CKSUM_OK;
+	  vlib_frame_no_append (f);
 	}
       n_left_to_next -= n_rx_packets;
       vlib_put_next_frame (vm, node, next_index, n_left_to_next);

src/plugins/memif/node.c

@@ -366,6 +366,7 @@ memif_device_input_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
 	  ef = vlib_frame_scalar_args (f);
 	  ef->sw_if_index = mif->sw_if_index;
 	  ef->hw_if_index = mif->hw_if_index;
+	  vlib_frame_no_append (f);
 	}
     }
 

src/scripts/generate_version_h

@@ -22,7 +22,7 @@ cat > ${1} << __EOF__
 #define VPP_BUILD_DATE "$(date)"
 #define VPP_BUILD_USER "$(whoami)"
 #define VPP_BUILD_HOST "$(hostname)"
-#define VPP_BUILD_TOPDIR "$(git rev-parse --show-toplevel)"
+#define VPP_BUILD_TOPDIR "$(git rev-parse --show-toplevel 2> /dev/null)"
 #define VPP_BUILD_VER "$(scripts/version)"
 #endif
 __EOF__

src/vcl/ldp.c

@@ -1700,6 +1700,11 @@ getsockopt (int fd, int level, int optname,
 	      else
 		rv = -EFAULT;
 	      break;
+	    case TCP_CONGESTION:
+	      strcpy (optval, "cubic");
+	      *optlen = strlen ("cubic");
+	      rv = 0;
+	      break;
 	    default:
 	      LDBG (0, "ERROR: fd %d: getsockopt SOL_TCP: sid %u, "
 		    "optname %d unsupported!", fd, vlsh, optname);
@@ -1808,6 +1813,10 @@ setsockopt (int fd, int level, int optname,
 	      rv = vls_attr (vlsh, VPPCOM_ATTR_SET_TCP_KEEPINTVL,
 			     (void *) optval, &optlen);
 	      break;
+	    case TCP_CONGESTION:
+	      /* Ignore */
+	      rv = 0;
+	      break;
 	    default:
 	      LDBG (0, "ERROR: fd %d: setsockopt() SOL_TCP: vlsh %u"
 		    "optname %d unsupported!", fd, vlsh, optname);

src/vlib/buffer_funcs.h

@@ -42,6 +42,10 @@
 
 #include <vppinfra/hash.h>
 #include <vppinfra/fifo.h>
+#include <vlib/buffer.h>
+#include <vlib/physmem_funcs.h>
+#include <vlib/main.h>
+#include <vlib/node.h>
 
 /** \file
     vlib buffer access methods.
@@ -1130,131 +1134,141 @@ vlib_validate_buffer_set_in_use (vlib_buffer_t * b, u32 expected)
 #endif
 }
 
-/** minimum data size of first buffer in a buffer chain */
-#define VLIB_BUFFER_CHAIN_MIN_FIRST_DATA_SIZE (256)
-
-/**
- * @brief compress buffer chain in a way where the first buffer is at least
- * VLIB_BUFFER_CHAIN_MIN_FIRST_DATA_SIZE long
- *
- * @param[in] vm - vlib_main
- * @param[in,out] first - first buffer in chain
- * @param[in,out] discard_vector - vector of buffer indexes which were removed
- * from the chain
- */
-always_inline void
-vlib_buffer_chain_compress (vlib_main_t * vm,
-			    vlib_buffer_t * first, u32 ** discard_vector)
+always_inline u32
+vlib_buffer_space_left_at_end (vlib_main_t * vm, vlib_buffer_t * b)
 {
-  if (first->current_length >= VLIB_BUFFER_CHAIN_MIN_FIRST_DATA_SIZE ||
-      !(first->flags & VLIB_BUFFER_NEXT_PRESENT))
-    {
-      /* this is already big enough or not a chain */
-      return;
-    }
-  /* probe free list to find allocated buffer size to avoid overfill */
-  vlib_buffer_free_list_index_t index;
-  vlib_buffer_free_list_t *free_list =
-    vlib_buffer_get_buffer_free_list (vm, first, &index);
-
-  u32 want_first_size = clib_min (VLIB_BUFFER_CHAIN_MIN_FIRST_DATA_SIZE,
-				  free_list->n_data_bytes -
-				  first->current_data);
-  do
-    {
-      vlib_buffer_t *second = vlib_get_buffer (vm, first->next_buffer);
-      u32 need = want_first_size - first->current_length;
-      u32 amount_to_copy = clib_min (need, second->current_length);
-      clib_memcpy_fast (((u8 *) vlib_buffer_get_current (first)) +
-			first->current_length,
-			vlib_buffer_get_current (second), amount_to_copy);
-      first->current_length += amount_to_copy;
-      second->current_data += amount_to_copy;
-      second->current_length -= amount_to_copy;
-      if (first->flags & VLIB_BUFFER_TOTAL_LENGTH_VALID)
-	{
-	  first->total_length_not_including_first_buffer -= amount_to_copy;
-	}
-      if (!second->current_length)
-	{
-	  vec_add1 (*discard_vector, first->next_buffer);
-	  if (second->flags & VLIB_BUFFER_NEXT_PRESENT)
-	    {
-	      first->next_buffer = second->next_buffer;
-	    }
-	  else
-	    {
-	      first->flags &= ~VLIB_BUFFER_NEXT_PRESENT;
-	    }
-	  second->flags &= ~VLIB_BUFFER_NEXT_PRESENT;
-	}
-    }
-  while ((first->current_length < want_first_size) &&
-	 (first->flags & VLIB_BUFFER_NEXT_PRESENT));
+  return b->data + VLIB_BUFFER_DATA_SIZE -
+    ((u8 *) vlib_buffer_get_current (b) + b->current_length);
 }
 
-/**
- * @brief linearize buffer chain - the first buffer is filled, if needed,
- * buffers are allocated and filled, returns free space in last buffer or
- * negative on failure
- *
- * @param[in] vm - vlib_main
- * @param[in,out] first - first buffer in chain
- */
-always_inline int
-vlib_buffer_chain_linearize (vlib_main_t * vm, vlib_buffer_t * first)
+always_inline u32
+vlib_buffer_chain_linearize (vlib_main_t * vm, vlib_buffer_t * b)
 {
-  vlib_buffer_t *b = first;
-  vlib_buffer_free_list_t *fl =
-    vlib_buffer_get_free_list (vm, vlib_buffer_get_free_list_index (b));
-  u32 buf_len = fl->n_data_bytes;
-  // free buffer chain starting from the second buffer
-  int free_count = (b->flags & VLIB_BUFFER_NEXT_PRESENT) != 0;
-  u32 chain_to_free = b->next_buffer;
+  vlib_buffer_t *db = b, *sb, *first = b;
+  int is_cloned = 0;
+  u32 bytes_left = 0, data_size;
+  u16 src_left, dst_left, n_buffers = 1;
+  u8 *dp, *sp;
+  u32 to_free = 0;
 
-  u32 len = vlib_buffer_length_in_chain (vm, b);
-  u32 free_len = buf_len - b->current_data - b->current_length;
-  int alloc_len = clib_max (len - free_len, 0);	//use the free len in the first buffer
-  int n_buffers = (alloc_len + buf_len - 1) / buf_len;
-  u32 new_buffers[n_buffers];
+  if (PREDICT_TRUE ((b->flags & VLIB_BUFFER_NEXT_PRESENT) == 0))
+    return 1;
 
-  u32 n_alloc = vlib_buffer_alloc (vm, new_buffers, n_buffers);
-  if (n_alloc != n_buffers)
+  data_size = VLIB_BUFFER_DATA_SIZE;
+
+  dst_left = vlib_buffer_space_left_at_end (vm, b);
+
+  while (b->flags & VLIB_BUFFER_NEXT_PRESENT)
     {
-      vlib_buffer_free_no_next (vm, new_buffers, n_alloc);
-      return -1;
+      b = vlib_get_buffer (vm, b->next_buffer);
+      if (b->n_add_refs > 0)
+	is_cloned = 1;
+      bytes_left += b->current_length;
+      n_buffers++;
     }
 
-  vlib_buffer_t *s = b;
-  while (s->flags & VLIB_BUFFER_NEXT_PRESENT)
+  /* if buffer is cloned, create completely new chain - unless everything fits
+   * into one buffer */
+  if (is_cloned && bytes_left >= dst_left)
     {
-      s = vlib_get_buffer (vm, s->next_buffer);
-      int d_free_len = buf_len - b->current_data - b->current_length;
-      ASSERT (d_free_len >= 0);
-      // chain buf and split write
-      u32 copy_len = clib_min (d_free_len, s->current_length);
-      u8 *d = vlib_buffer_put_uninit (b, copy_len);
-      clib_memcpy (d, vlib_buffer_get_current (s), copy_len);
-      int rest = s->current_length - copy_len;
-      if (rest > 0)
+      u32 len = 0;
+      u32 space_needed = bytes_left - dst_left;
+      u32 tail;
+
+      if (vlib_buffer_alloc (vm, &tail, 1) == 0)
+	return 0;
+
+      ++n_buffers;
+      len += data_size;
+      b = vlib_get_buffer (vm, tail);
+
+      while (len < space_needed)
 	{
-	  //prev buf is full
-	  ASSERT (vlib_buffer_get_tail (b) == b->data + buf_len);
-	  ASSERT (n_buffers > 0);
-	  b = vlib_buffer_chain_buffer (vm, b, new_buffers[--n_buffers]);
-	  //make full use of the new buffers
-	  b->current_data = 0;
-	  d = vlib_buffer_put_uninit (b, rest);
-	  clib_memcpy (d, vlib_buffer_get_current (s) + copy_len, rest);
+	  u32 bi;
+	  if (vlib_buffer_alloc (vm, &bi, 1) == 0)
+	    {
+	      vlib_buffer_free_one (vm, tail);
+	      return 0;
+	    }
+	  b->flags = VLIB_BUFFER_NEXT_PRESENT;
+	  b->next_buffer = bi;
+	  b = vlib_get_buffer (vm, bi);
+	  len += data_size;
+	  n_buffers++;
+	}
+      sb = vlib_get_buffer (vm, first->next_buffer);
+      to_free = first->next_buffer;
+      first->next_buffer = tail;
+    }
+  else
+    sb = vlib_get_buffer (vm, first->next_buffer);
+
+  src_left = sb->current_length;
+  sp = vlib_buffer_get_current (sb);
+  dp = vlib_buffer_get_tail (db);
+
+  while (bytes_left)
+    {
+      u16 bytes_to_copy;
+
+      if (dst_left == 0)
+	{
+	  if (db != first)
+	    db->current_data = 0;
+	  db->current_length = dp - (u8 *) vlib_buffer_get_current (db);
+	  ASSERT (db->flags & VLIB_BUFFER_NEXT_PRESENT);
+	  db = vlib_get_buffer (vm, db->next_buffer);
+	  dst_left = data_size;
+	  dp = db->data;
+	}
+
+      while (src_left == 0)
+	{
+	  ASSERT (sb->flags & VLIB_BUFFER_NEXT_PRESENT);
+	  sb = vlib_get_buffer (vm, sb->next_buffer);
+	  src_left = sb->current_length;
+	  sp = vlib_buffer_get_current (sb);
+	}
+
+      bytes_to_copy = clib_min (dst_left, src_left);
+
+      if (dp != sp)
+	{
+	  if (sb == db)
+	    bytes_to_copy = clib_min (bytes_to_copy, sp - dp);
+
+	  clib_memcpy_fast (dp, sp, bytes_to_copy);
+	}
+
+      src_left -= bytes_to_copy;
+      dst_left -= bytes_to_copy;
+      dp += bytes_to_copy;
+      sp += bytes_to_copy;
+      bytes_left -= bytes_to_copy;
+    }
+  if (db != first)
+    db->current_data = 0;
+  db->current_length = dp - (u8 *) vlib_buffer_get_current (db);
+
+  if (is_cloned && to_free)
+    vlib_buffer_free_one (vm, to_free);
+  else
+    {
+      if (db->flags & VLIB_BUFFER_NEXT_PRESENT)
+	vlib_buffer_free_one (vm, db->next_buffer);
+      db->flags &= ~VLIB_BUFFER_NEXT_PRESENT;
+      b = first;
+      n_buffers = 1;
+      while (b->flags & VLIB_BUFFER_NEXT_PRESENT)
+	{
+	  b = vlib_get_buffer (vm, b->next_buffer);
+	  ++n_buffers;
 	}
     }
-  vlib_buffer_free (vm, &chain_to_free, free_count);
-  b->flags &= ~VLIB_BUFFER_TOTAL_LENGTH_VALID;
-  if (b == first)		/* no buffers addeed */
-    b->flags &= ~VLIB_BUFFER_NEXT_PRESENT;
-  ASSERT (len == vlib_buffer_length_in_chain (vm, first));
-  ASSERT (n_buffers == 0);
-  return buf_len - b->current_data - b->current_length;
+
+  first->flags &= ~VLIB_BUFFER_TOTAL_LENGTH_VALID;
+
+  return n_buffers;
 }
 
 #endif /* included_vlib_buffer_funcs_h */

src/vlib/main.c

@@ -233,7 +233,7 @@ vlib_frame_free (vlib_main_t * vm, vlib_node_runtime_t * r, vlib_frame_t * f)
 	ASSERT (nf->frame_index != frame_index);
     }
 
-  f->frame_flags &= ~VLIB_FRAME_IS_ALLOCATED;
+  f->frame_flags &= ~(VLIB_FRAME_IS_ALLOCATED | VLIB_FRAME_NO_APPEND);
 
   vec_add1 (fs->free_frame_indices, frame_index);
   ASSERT (fs->n_alloc_frames > 0);
@@ -387,9 +387,11 @@ vlib_get_next_frame_internal (vlib_main_t * vm,
       f->flags = 0;
     }
 
-  /* Allocate new frame if current one is already full. */
+  /* Allocate new frame if current one is marked as no-append or
+     it is already full. */
   n_used = f->n_vectors;
-  if (n_used >= VLIB_FRAME_SIZE || (allocate_new_next_frame && n_used > 0))
+  if (n_used >= VLIB_FRAME_SIZE || (allocate_new_next_frame && n_used > 0) ||
+      (f->frame_flags & VLIB_FRAME_NO_APPEND))
     {
       /* Old frame may need to be freed after dispatch, since we'll have
          two redundant frames from node -> next node. */
@@ -1347,7 +1349,7 @@ dispatch_pending_node (vlib_main_t * vm, uword pending_frame_index,
 				   VLIB_NODE_STATE_POLLING,
 				   f, last_time_stamp);
 
-  f->frame_flags &= ~VLIB_FRAME_PENDING;
+  f->frame_flags &= ~(VLIB_FRAME_PENDING | VLIB_FRAME_NO_APPEND);
 
   /* Frame is ready to be used again, so restore it. */
   if (restore_frame_index != ~0)

src/vlib/node.h

@@ -438,6 +438,9 @@ typedef struct
 #define VLIB_FRAME_NO_FREE_AFTER_DISPATCH \
   VLIB_NODE_FLAG_FRAME_NO_FREE_AFTER_DISPATCH
 
+  /* Don't append this frame */
+#define VLIB_FRAME_NO_APPEND (1 << 14)
+
   /* This next frame owns enqueue to node
      corresponding to node_runtime_index. */
 #define VLIB_FRAME_OWNER (1 << 15)

src/vlib/node_funcs.h

@@ -242,6 +242,12 @@ vlib_get_frame (vlib_main_t * vm, uword frame_index)
   return f;
 }
 
+always_inline void
+vlib_frame_no_append (vlib_frame_t * f)
+{
+  f->frame_flags |= VLIB_FRAME_NO_APPEND;
+}
+
 always_inline u32
 vlib_frame_index (vlib_main_t * vm, vlib_frame_t * f)
 {

src/vlib/physmem.c

@@ -29,6 +29,15 @@
 #include <vlib/pci/pci.h>
 #include <vlib/linux/vfio.h>
 
+#ifdef __x86_64__
+/* we keep physmem in low 38 bits of VA address space as some
+   IOMMU implamentation cannot map above that range */
+#define VLIB_PHYSMEM_DEFAULT_BASE_ADDDR		(1ULL << 36)
+#else
+/* let kernel decide */
+#define VLIB_PHYSMEM_DEFAULT_BASE_ADDDR		0
+#endif
+
 clib_error_t *
 vlib_physmem_shared_map_create (vlib_main_t * vm, char *name, uword size,
 				u32 log2_page_sz, u32 numa_node,
@@ -102,7 +111,11 @@ vlib_physmem_init (vlib_main_t * vm)
 			      CLIB_CACHE_LINE_BYTES);
   memset (p, 0, sizeof (clib_pmalloc_main_t));
   vpm->pmalloc_main = (clib_pmalloc_main_t *) p;
-  clib_pmalloc_init (vpm->pmalloc_main, 0);
+
+  if (vpm->base_addr == 0)
+    vpm->base_addr = VLIB_PHYSMEM_DEFAULT_BASE_ADDDR;
+
+  clib_pmalloc_init (vpm->pmalloc_main, vpm->base_addr, 0);
 
   return error;
 }
@@ -151,6 +164,25 @@ VLIB_CLI_COMMAND (show_physmem_command, static) = {
 };
 /* *INDENT-ON* */
 
+static clib_error_t *
+vlib_physmem_config (vlib_main_t * vm, unformat_input_t * input)
+{
+  vlib_physmem_main_t *vpm = &vm->physmem_main;
+
+  while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
+    {
+      if (unformat (input, "base-addr 0x%lx", &vpm->base_addr))
+	;
+      else
+	return unformat_parse_error (input);
+    }
+
+  unformat_free (input);
+  return 0;
+}
+
+VLIB_EARLY_CONFIG_FUNCTION (vlib_physmem_config, "physmem");
+
 /*
  * fd.io coding-style-patch-verification: ON
  *

src/vlib/physmem.h

@@ -56,6 +56,7 @@ typedef struct
 typedef struct
 {
   u32 flags;
+  uword base_addr;
 #define VLIB_PHYSMEM_MAIN_F_HAVE_PAGEMAP	(1 << 0)
 #define VLIB_PHYSMEM_MAIN_F_HAVE_IOMMU		(1 << 1)
   vlib_physmem_map_t *maps;

src/vlib/physmem_funcs.h

@@ -40,6 +40,11 @@
 #ifndef included_vlib_physmem_funcs_h
 #define included_vlib_physmem_funcs_h
 
+#include <vppinfra/clib.h>
+#include <vppinfra/clib_error.h>
+#include <vlib/physmem.h>
+#include <vlib/main.h>
+
 clib_error_t *vlib_physmem_init (vlib_main_t * vm);
 clib_error_t *vlib_physmem_shared_map_create (vlib_main_t * vm, char *name,
 					      uword size, u32 log2_page_sz,

src/vlib/threads.c

@@ -1360,7 +1360,7 @@ vlib_worker_thread_fork_fixup (vlib_fork_fixup_t which)
 #endif
 
 void
-vlib_worker_thread_barrier_sync_int (vlib_main_t * vm)
+vlib_worker_thread_barrier_sync_int (vlib_main_t * vm, const char *func_name)
 {
   f64 deadline;
   f64 now;
@@ -1374,6 +1374,7 @@ vlib_worker_thread_barrier_sync_int (vlib_main_t * vm)
 
   ASSERT (vlib_get_thread_index () == 0);
 
+  vlib_worker_threads[0].barrier_caller = func_name;
   count = vec_len (vlib_mains) - 1;
 
   /* Record entry relative to last close */

src/vlib/threads.h

@@ -201,9 +201,10 @@ u32 vlib_frame_queue_main_init (u32 node_index, u32 frame_queue_nelts);
 #define BARRIER_SYNC_TIMEOUT (1.0)
 #endif
 
-#define vlib_worker_thread_barrier_sync(X) {vlib_worker_threads[0].barrier_caller=__FUNCTION__;vlib_worker_thread_barrier_sync_int(X);}
+#define vlib_worker_thread_barrier_sync(X) {vlib_worker_thread_barrier_sync_int(X, __FUNCTION__);}
 
-void vlib_worker_thread_barrier_sync_int (vlib_main_t * vm);
+void vlib_worker_thread_barrier_sync_int (vlib_main_t * vm,
+					  const char *func_name);
 void vlib_worker_thread_barrier_release (vlib_main_t * vm);
 void vlib_worker_thread_node_refork (void);
 

src/vnet/bonding/cli.c

@@ -531,11 +531,13 @@ 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/device.c

@@ -104,17 +104,6 @@ bond_set_l2_mode_function (vnet_main_t * vnm,
 	ethernet_set_rx_redirect (vnm, sif_hw, 1);
       }
     }
-  else if ((bif_hw->l2_if_count == 0) && (l2_if_adjust == -1))
-    {
-      /* Just removed last L2 subinterface on this port */
-      vec_foreach (sw_if_index, bif->slaves)
-      {
-	sif_hw = vnet_get_sup_hw_interface (vnm, *sw_if_index);
-
-	/* Allow ip packets to go directly to ip4-input etc */
-	ethernet_set_rx_redirect (vnm, sif_hw, 0);
-      }
-    }
 
   return 0;
 }

src/vnet/bonding/node.c

@@ -396,19 +396,21 @@ 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)
-	    {
-	      bond_enable_collecting_distributing (vm, sif);
-	    }
+	  vnet_main_t *vnm = vnet_get_main ();
+	  vnet_hw_interface_t *hw =
+	    vnet_get_sup_hw_interface (vnm, sw_if_index);
+
+	  if (hw->flags & VNET_HW_INTERFACE_FLAG_LINK_UP)
+	    bond_enable_collecting_distributing (vm, sif);
 	}
     }
 
@@ -429,19 +431,16 @@ 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

@@ -250,6 +250,14 @@ 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
@@ -293,9 +301,6 @@ 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

@@ -244,6 +244,9 @@ 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/devices/virtio/vhost_user_input.c

@@ -402,6 +402,7 @@ vhost_user_if_input (vlib_main_t * vm,
       ef = vlib_frame_scalar_args (f);
       ef->sw_if_index = vui->sw_if_index;
       ef->hw_if_index = vui->hw_if_index;
+      vlib_frame_no_append (f);
     }
 
   while (n_left > 0)