Introduce l{2,3,4}_hdr_offset fields in the buffer metadata

To save space in the first cacheline following is changed:

- total_length_not_including_first_buffer moved to the 2nd cacheline.
This field is used only when VLIB_BUFFER_TOTAL_LENGTH_VALID and
VLIB_BUFFER_NEXT_PRESENT are both set.

- free_list_index is now stored in 4bits inside flags, which
allows up to 16 free lists. In case we need more we can store index
in the 2nd cachelin

Change-Id: Ic8521350819391af470d31d3fa1013e67ecb7681
Signed-off-by: Damjan Marion <damarion@cisco.com>

3 files changed, 68 insertions, 38 deletions

diff --git a/src/vlib/buffer.c b/src/vlib/buffer.c
index b2a095cfb1a..53b60c165c1 100644
--- a/src/vlib/buffer.c
+++ b/src/vlib/buffer.c
@@ -72,8 +72,8 @@ format_vlib_buffer (u8 * s, va_list * args)
   uword indent = format_get_indent (s);
 
   s = format (s, "current data %d, length %d, free-list %d, clone-count %u",
-	      b->current_data, b->current_length, b->free_list_index,
-	      b->n_add_refs);
+	      b->current_data, b->current_length,
+	      vlib_buffer_get_free_list_index (b), b->n_add_refs);
 
   if (b->flags & VLIB_BUFFER_TOTAL_LENGTH_VALID)
     s = format (s, ", totlen-nifb %d",
@@ -163,10 +163,14 @@ vlib_validate_buffer_helper (vlib_main_t * vm,
   vlib_buffer_main_t *bm = vm->buffer_main;
   vlib_buffer_free_list_t *fl;
 
-  if (pool_is_free_index (bm->buffer_free_list_pool, b->free_list_index))
-    return format (0, "unknown free list 0x%x", b->free_list_index);
+  if (pool_is_free_index
+      (bm->buffer_free_list_pool, vlib_buffer_get_free_list_index (b)))
+    return format (0, "unknown free list 0x%x",
+		   vlib_buffer_get_free_list_index (b));
 
-  fl = pool_elt_at_index (bm->buffer_free_list_pool, b->free_list_index);
+  fl =
+    pool_elt_at_index (bm->buffer_free_list_pool,
+		       vlib_buffer_get_free_list_index (b));
 
   if ((signed) b->current_data < (signed) -VLIB_BUFFER_PRE_DATA_SIZE)
     return format (0, "current data %d before pre-data", b->current_data);
@@ -388,7 +392,7 @@ vlib_buffer_create_free_list_helper (vlib_main_t * vm,
   f->name = clib_mem_is_vec (name) ? name : format (0, "%s", name);
 
   /* Setup free buffer template. */
-  f->buffer_init_template.free_list_index = f->index;
+  vlib_buffer_set_free_list_index (&f->buffer_init_template, f->index);
   f->buffer_init_template.n_add_refs = 0;
 
   if (is_public)
diff --git a/src/vlib/buffer.h b/src/vlib/buffer.h
index b20538b7f8c..c810db4e860 100644
--- a/src/vlib/buffer.h
+++ b/src/vlib/buffer.h
@@ -72,6 +72,7 @@ typedef struct
                           the end of this buffer.
                        */
   u32 flags; /**< buffer flags:
+                <br> VLIB_BUFFER_FREE_LIST_INDEX_MASK: bits used to store free list index,
                 <br> VLIB_BUFFER_IS_TRACED: trace this buffer.
                 <br> VLIB_BUFFER_NEXT_PRESENT: this is a multi-chunk buffer.
                 <br> VLIB_BUFFER_TOTAL_LENGTH_VALID: as it says
@@ -82,28 +83,26 @@ typedef struct
                 set to avoid adding it to a flow report
                 <br> VLIB_BUFFER_FLAG_USER(n): user-defined bit N
              */
-#define VLIB_BUFFER_IS_TRACED (1 << 0)
-#define VLIB_BUFFER_LOG2_NEXT_PRESENT (1)
+
+/* any change to the following line requres update of
+ * vlib_buffer_get_free_list_index(...) and
+ * vlib_buffer_set_free_list_index(...) functions */
+#define VLIB_BUFFER_FREE_LIST_INDEX_MASK ((1 << 4) - 1)
+
+#define VLIB_BUFFER_IS_TRACED (1 << 4)
+#define VLIB_BUFFER_LOG2_NEXT_PRESENT (5)
 #define VLIB_BUFFER_NEXT_PRESENT (1 << VLIB_BUFFER_LOG2_NEXT_PRESENT)
-#define VLIB_BUFFER_IS_RECYCLED (1 << 2)
-#define VLIB_BUFFER_TOTAL_LENGTH_VALID (1 << 3)
-#define VLIB_BUFFER_REPL_FAIL (1 << 4)
-#define VLIB_BUFFER_RECYCLE (1 << 5)
-#define VLIB_BUFFER_FLOW_REPORT (1 << 6)
-#define VLIB_BUFFER_EXT_HDR_VALID (1 << 7)
+#define VLIB_BUFFER_IS_RECYCLED (1 << 6)
+#define VLIB_BUFFER_TOTAL_LENGTH_VALID (1 << 7)
+#define VLIB_BUFFER_REPL_FAIL (1 << 8)
+#define VLIB_BUFFER_RECYCLE (1 << 9)
+#define VLIB_BUFFER_FLOW_REPORT (1 << 10)
+#define VLIB_BUFFER_EXT_HDR_VALID (1 << 11)
 
   /* User defined buffer flags. */
 #define LOG2_VLIB_BUFFER_FLAG_USER(n) (32 - (n))
 #define VLIB_BUFFER_FLAG_USER(n) (1 << LOG2_VLIB_BUFFER_FLAG_USER(n))
 
-  u32 free_list_index; /**< Buffer free list that this buffer was
-                          allocated from and will be freed to.
-                       */
-
-  u32 total_length_not_including_first_buffer;
-  /**< Only valid for first buffer in chain. Current length plus
-     total length given here give total number of bytes in buffer chain.
-  */
     STRUCT_MARK (template_end);
 
   u32 next_buffer;   /**< Next buffer for this linked-list of buffers.
@@ -128,7 +127,7 @@ typedef struct
                           Before allocating any of it, discussion required!
                        */
 
-  u32 opaque[8]; /**< Opaque data used by sub-graphs for their own purposes.
+  u32 opaque[10]; /**< Opaque data used by sub-graphs for their own purposes.
                     See .../vnet/vnet/buffer.h
                  */
     CLIB_CACHE_LINE_ALIGN_MARK (cacheline1);
@@ -137,7 +136,12 @@ typedef struct
                       if VLIB_PACKET_IS_TRACED flag is set.
                    */
   u32 recycle_count; /**< Used by L2 path recycle code */
-  u32 opaque2[14];  /**< More opaque data, currently unused */
+
+  u32 total_length_not_including_first_buffer;
+  /**< Only valid for first buffer in chain. Current length plus
+     total length given here give total number of bytes in buffer chain.
+  */
+  u32 opaque2[13];  /**< More opaque data, currently unused */
 
   /***** end of second cache line */
     CLIB_CACHE_LINE_ALIGN_MARK (cacheline2);
diff --git a/src/vlib/buffer_funcs.h b/src/vlib/buffer_funcs.h
index 97442e12239..1aaac0b2a11 100644
--- a/src/vlib/buffer_funcs.h
+++ b/src/vlib/buffer_funcs.h
@@ -106,12 +106,15 @@ uword vlib_buffer_length_in_chain_slow_path (vlib_main_t * vm,
 always_inline uword
 vlib_buffer_length_in_chain (vlib_main_t * vm, vlib_buffer_t * b)
 {
-  uword l = b->current_length + b->total_length_not_including_first_buffer;
-  if (PREDICT_FALSE ((b->flags & (VLIB_BUFFER_NEXT_PRESENT
-				  | VLIB_BUFFER_TOTAL_LENGTH_VALID))
-		     == VLIB_BUFFER_NEXT_PRESENT))
-    return vlib_buffer_length_in_chain_slow_path (vm, b);
-  return l;
+  uword len = b->current_length;
+
+  if (PREDICT_TRUE ((b->flags & VLIB_BUFFER_NEXT_PRESENT) == 0))
+    return len;
+
+  if (PREDICT_TRUE (b->flags & VLIB_BUFFER_TOTAL_LENGTH_VALID))
+    return len + b->total_length_not_including_first_buffer;
+
+  return vlib_buffer_length_in_chain_slow_path (vm, b);
 }
 
 /** \brief Get length in bytes of the buffer index buffer chain
@@ -261,6 +264,24 @@ vlib_buffer_round_size (u32 size)
   return round_pow2 (size, sizeof (vlib_buffer_t));
 }
 
+always_inline u32
+vlib_buffer_get_free_list_index (vlib_buffer_t * b)
+{
+  return b->flags & VLIB_BUFFER_FREE_LIST_INDEX_MASK;
+}
+
+always_inline void
+vlib_buffer_set_free_list_index (vlib_buffer_t * b, u32 index)
+{
+  /* if there is an need for more free lists we should consider
+     storig data in the 2nd cacheline */
+  ASSERT (VLIB_BUFFER_FREE_LIST_INDEX_MASK & 1);
+  ASSERT (index <= VLIB_BUFFER_FREE_LIST_INDEX_MASK);
+
+  b->flags &= ~VLIB_BUFFER_FREE_LIST_INDEX_MASK;
+  b->flags |= index & VLIB_BUFFER_FREE_LIST_INDEX_MASK;
+}
+
 /** \brief Allocate buffers from specific freelist into supplied array
 
     @param vm - (vlib_main_t *) vlib main data structure pointer
@@ -381,7 +402,7 @@ vlib_buffer_get_buffer_free_list (vlib_main_t * vm, vlib_buffer_t * b,
   vlib_buffer_main_t *bm = vm->buffer_main;
   u32 i;
 
-  *index = i = b->free_list_index;
+  *index = i = vlib_buffer_get_free_list_index (b);
   return pool_elt_at_index (bm->buffer_free_list_pool, i);
 }
 
@@ -569,7 +590,8 @@ vlib_buffer_clone (vlib_main_t * vm, u32 src_buffer, u32 * buffers,
     }
 
   n_buffers = vlib_buffer_alloc_from_free_list (vm, buffers, n_buffers,
-						s->free_list_index);
+						vlib_buffer_get_free_list_index
+						(s));
   if (PREDICT_FALSE (n_buffers == 0))
     {
       buffers[0] = src_buffer;
@@ -581,7 +603,8 @@ vlib_buffer_clone (vlib_main_t * vm, u32 src_buffer, u32 * buffers,
       vlib_buffer_t *d = vlib_get_buffer (vm, buffers[i]);
       d->current_data = s->current_data;
       d->current_length = head_end_offset;
-      d->free_list_index = s->free_list_index;
+      vlib_buffer_set_free_list_index (d,
+				       vlib_buffer_get_free_list_index (s));
       d->total_length_not_including_first_buffer =
 	s->total_length_not_including_first_buffer + s->current_length -
 	head_end_offset;
@@ -615,7 +638,8 @@ vlib_buffer_attach_clone (vlib_main_t * vm, vlib_buffer_t * head,
 			  vlib_buffer_t * tail)
 {
   ASSERT ((head->flags & VLIB_BUFFER_NEXT_PRESENT) == 0);
-  ASSERT (head->free_list_index == tail->free_list_index);
+  ASSERT (vlib_buffer_get_free_list_index (head) ==
+	  vlib_buffer_get_free_list_index (tail));
 
   head->flags |= VLIB_BUFFER_NEXT_PRESENT;
   head->flags &= ~VLIB_BUFFER_TOTAL_LENGTH_VALID;
@@ -791,7 +815,7 @@ vlib_buffer_init_for_free_list (vlib_buffer_t * dst,
 	  CLIB_CACHE_LINE_BYTES * 2);
 
   /* Make sure buffer template is sane. */
-  ASSERT (fl->index == fl->buffer_init_template.free_list_index);
+  ASSERT (fl->index == vlib_buffer_get_free_list_index (src));
 
   clib_memcpy (STRUCT_MARK_PTR (dst, template_start),
 	       STRUCT_MARK_PTR (src, template_start),
@@ -806,7 +830,6 @@ vlib_buffer_init_for_free_list (vlib_buffer_t * dst,
   _(current_data);
   _(current_length);
   _(flags);
-  _(free_list_index);
 #undef _
   ASSERT (dst->total_length_not_including_first_buffer == 0);
   ASSERT (dst->n_add_refs == 0);
@@ -832,7 +855,7 @@ vlib_buffer_init_two_for_free_list (vlib_buffer_t * dst0,
   vlib_buffer_t *src = &fl->buffer_init_template;
 
   /* Make sure buffer template is sane. */
-  ASSERT (fl->index == fl->buffer_init_template.free_list_index);
+  ASSERT (fl->index == vlib_buffer_get_free_list_index (src));
 
   clib_memcpy (STRUCT_MARK_PTR (dst0, template_start),
 	       STRUCT_MARK_PTR (src, template_start),
@@ -853,7 +876,6 @@ vlib_buffer_init_two_for_free_list (vlib_buffer_t * dst0,
   _(current_data);
   _(current_length);
   _(flags);
-  _(free_list_index);
 #undef _
 
   ASSERT (dst0->total_length_not_including_first_buffer == 0);