dpdk: initialize rte_mbuf during mempool dequeue

In some cases it may happen that buffer is allocated by DPDK, and freed
by VPP native code. In such cases dpdk metadata is not reset, so we need
to do that during mempool dequeue. Template approach is taken to reduce
cost of that operation.

Type: fix
Fixes: 910d369

Change-Id: Ic239007cfc8fbceb965021c56963cda9d53f63be
Signed-off-by: Damjan Marion <damarion@cisco.com>

2 files changed, 47 insertions, 1 deletions

diff --git a/src/plugins/dpdk/buffer.c b/src/plugins/dpdk/buffer.c
index dd7bf4cc9d1..411d2ddc91c 100644
--- a/src/plugins/dpdk/buffer.c
+++ b/src/plugins/dpdk/buffer.c
@@ -27,9 +27,11 @@
 STATIC_ASSERT (VLIB_BUFFER_PRE_DATA_SIZE == RTE_PKTMBUF_HEADROOM,
 	       "VLIB_BUFFER_PRE_DATA_SIZE must be equal to RTE_PKTMBUF_HEADROOM");
 
+extern struct rte_mbuf *dpdk_mbuf_template_by_pool_index;
 #ifndef CLIB_MARCH_VARIANT
 struct rte_mempool **dpdk_mempool_by_buffer_pool_index = 0;
 struct rte_mempool **dpdk_no_cache_mempool_by_buffer_pool_index = 0;
+struct rte_mbuf *dpdk_mbuf_template_by_pool_index = 0;
 
 clib_error_t *
 dpdk_buffer_pool_init (vlib_main_t * vm, vlib_buffer_pool_t * bp)
@@ -102,6 +104,14 @@ dpdk_buffer_pool_init (vlib_main_t * vm, vlib_buffer_pool_t * bp)
   /* call the object initializers */
   rte_mempool_obj_iter (mp, rte_pktmbuf_init, 0);
 
+  /* create mbuf header tempate from the first buffer in the pool */
+  vec_validate_aligned (dpdk_mbuf_template_by_pool_index, bp->index,
+			CLIB_CACHE_LINE_BYTES);
+  clib_memcpy (vec_elt_at_index (dpdk_mbuf_template_by_pool_index, bp->index),
+	       rte_mbuf_from_vlib_buffer (vlib_buffer_ptr_from_index
+					  (buffer_mem_start, *bp->buffers,
+					   0)), sizeof (struct rte_mbuf));
+
   /* *INDENT-OFF* */
   vec_foreach (bi, bp->buffers)
     {
@@ -241,7 +251,6 @@ dpdk_ops_vpp_enqueue_no_cache_one (vlib_main_t * vm, struct rte_mempool *old,
   if (clib_atomic_sub_fetch (&b->ref_count, 1) == 0)
     {
       u32 bi = vlib_get_buffer_index (vm, b);
-      mb->pool = new;
       vlib_buffer_copy_template (b, bt);
       vlib_buffer_pool_put (vm, bt->buffer_pool_index, &bi, 1);
       return;
@@ -283,6 +292,37 @@ CLIB_MULTIARCH_FN (dpdk_ops_vpp_enqueue_no_cache) (struct rte_mempool * cmp,
 
 CLIB_MARCH_FN_REGISTRATION (dpdk_ops_vpp_enqueue_no_cache);
 
+static_always_inline void
+dpdk_mbuf_init_from_template (struct rte_mbuf **mba, struct rte_mbuf *mt,
+			      int count)
+{
+  /* Assumptions about rte_mbuf layout */
+  STATIC_ASSERT_OFFSET_OF (struct rte_mbuf, buf_addr, 0);
+  STATIC_ASSERT_OFFSET_OF (struct rte_mbuf, buf_iova, 8);
+  STATIC_ASSERT_SIZEOF_ELT (struct rte_mbuf, buf_iova, 8);
+  STATIC_ASSERT_SIZEOF_ELT (struct rte_mbuf, buf_iova, 8);
+  STATIC_ASSERT_SIZEOF (struct rte_mbuf, 128);
+
+  while (count--)
+    {
+      struct rte_mbuf *mb = mba[0];
+      int i;
+      /* bytes 0 .. 15 hold buf_addr and buf_iova which we need to preserve */
+      /* copy bytes 16 .. 31 */
+      *((u8x16 *) mb + 1) = *((u8x16 *) mt + 1);
+
+      /* copy bytes 32 .. 127 */
+#ifdef CLIB_HAVE_VEC256
+      for (i = 1; i < 4; i++)
+	*((u8x32 *) mb + i) = *((u8x32 *) mt + i);
+#else
+      for (i = 2; i < 8; i++)
+	*((u8x16 *) mb + i) = *((u8x16 *) mt + i);
+#endif
+      mba++;
+    }
+}
+
 int
 CLIB_MULTIARCH_FN (dpdk_ops_vpp_dequeue) (struct rte_mempool * mp,
 					  void **obj_table, unsigned n)
@@ -292,6 +332,7 @@ CLIB_MULTIARCH_FN (dpdk_ops_vpp_dequeue) (struct rte_mempool * mp,
   u32 bufs[batch_size], total = 0, n_alloc = 0;
   u8 buffer_pool_index = mp->pool_id;
   void **obj = obj_table;
+  struct rte_mbuf t = dpdk_mbuf_template_by_pool_index[buffer_pool_index];
 
   while (n >= batch_size)
     {
@@ -302,6 +343,7 @@ CLIB_MULTIARCH_FN (dpdk_ops_vpp_dequeue) (struct rte_mempool * mp,
 
       vlib_get_buffers_with_offset (vm, bufs, obj, batch_size,
 				    -(i32) sizeof (struct rte_mbuf));
+      dpdk_mbuf_init_from_template ((struct rte_mbuf **) obj, &t, batch_size);
       total += batch_size;
       obj += batch_size;
       n -= batch_size;
@@ -316,6 +358,7 @@ CLIB_MULTIARCH_FN (dpdk_ops_vpp_dequeue) (struct rte_mempool * mp,
 
       vlib_get_buffers_with_offset (vm, bufs, obj, n,
 				    -(i32) sizeof (struct rte_mbuf));
+      dpdk_mbuf_init_from_template ((struct rte_mbuf **) obj, &t, n);
     }
 
   return 0;
diff --git a/src/vppinfra/error_bootstrap.h b/src/vppinfra/error_bootstrap.h
index 18d2176812d..1cd91058574 100644
--- a/src/vppinfra/error_bootstrap.h
+++ b/src/vppinfra/error_bootstrap.h
@@ -88,6 +88,9 @@ do {							\
 #define STATIC_ASSERT_SIZEOF(d, s) \
   STATIC_ASSERT (sizeof (d) == s, "Size of " #d " must be " # s " bytes")
 
+#define STATIC_ASSERT_SIZEOF_ELT(d, e, s) \
+  STATIC_ASSERT (sizeof (((d *)0)->e) == s, "Size of " #d "." #e " must be " # s " bytes")
+
 #define STATIC_ASSERT_OFFSET_OF(s, e, o) \
   STATIC_ASSERT (STRUCT_OFFSET_OF(s,e) == o, "Offset of " #s "." #e " must be " # o)